Add files using upload-large-folder tool

CITATION.cff

@@ -0,0 +1,40 @@
+cff-version: 1.2.0
+message: "If you use VoxMemBench, please cite it as below."
+title: "VoxMemBench: a benchmark for long-term spoken conversational memory"
+abstract: >-
+  VoxMemBench measures whether a model can recall evidence from many
+  time-separated sessions of a spoken conversation. Evidence is carried by
+  speech semantics, speaker identity, paralinguistic delivery or environmental
+  sound, and each question is posed at four context lengths over a nested
+  history.
+type: dataset
+version: 1.0.0
+license: CC-BY-NC-4.0
+license-url: "LICENSE"
+keywords:
+  - spoken dialogue
+  - long-context
+  - conversational memory
+  - paralinguistics
+  - environmental sound
+authors:
+  - name: "The VoxMemBench authors"
+references:
+  - type: data
+    title: "CSTR VCTK Corpus (version 0.92)"
+    authors:
+      - family-names: Yamagishi
+        given-names: Junichi
+      - family-names: Veaux
+        given-names: Christophe
+      - family-names: MacDonald
+        given-names: Kirsten
+    doi: "10.7488/ds/2645"
+    license: CC-BY-4.0
+  - type: data
+    title: "ESC-50: Dataset for Environmental Sound Classification"
+    authors:
+      - family-names: Piczak
+        given-names: "Karol J."
+    doi: "10.7910/DVN/YDEPUT"
+    year: 2015

LICENSE

@@ -0,0 +1,35 @@
+VoxMemBench
+Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
+
+This dataset is licensed under CC BY-NC 4.0.
+https://creativecommons.org/licenses/by-nc/4.0/
+
+You may share and adapt it for non-commercial purposes, with attribution.
+
+
+Sources and required attribution
+--------------------------------
+
+Speech
+    The user speech in this benchmark is synthetic, voice-cloned from
+    reference speakers in the CSTR VCTK Corpus (version 0.92), Yamagishi,
+    Veaux and MacDonald, University of Edinburgh,
+    https://doi.org/10.7488/ds/2645, licensed CC BY 4.0.
+    The 30 reference speakers used are listed in metadata/speakers.json.
+
+Environmental sound
+    Environmental audio is mixed from clips in ESC-50, Piczak, K. J. (2015),
+    "ESC: Dataset for Environmental Sound Classification",
+    https://doi.org/10.7910/DVN/YDEPUT. ESC-50 as a whole is CC BY-NC 3.0;
+    its clips carry individual Freesound licences, and the 45 clips used are
+    listed with their licence in metadata/environmental_sources.json.
+
+The non-commercial term of this release follows from ESC-50: five of the clips
+used are CC BY-NC 3.0 and three are CC Sampling Plus 1.0, so no part of the
+benchmark is offered for commercial use.
+
+
+No warranty
+-----------
+
+The dataset is provided as is, without warranty of any kind.

README.md

@@ -0,0 +1,323 @@
+---
+license: cc-by-nc-4.0
+language:
+  - en
+task_categories:
+  - question-answering
+  - audio-text-to-text
+pretty_name: VoxMemBench
+tags:
+  - benchmark
+  - memory
+  - long-context
+  - spoken-dialogue
+  - conversational-memory
+  - paralinguistics
+  - speaker-recognition
+  - environmental-sound
+size_categories:
+  - 1K<n<10K
+configs:
+  - config_name: 8k
+    data_files: data/8k/*/*.parquet
+  - config_name: 16k
+    data_files: data/16k/*/*.parquet
+  - config_name: 32k
+    data_files: data/32k/*/*.parquet
+    default: true
+  - config_name: 64k
+    data_files: data/64k/*/*.parquet
+  - config_name: 8k_speech_semantics
+    data_files: data/8k/speech_semantics/*.parquet
+  - config_name: 8k_speaker_information
+    data_files: data/8k/speaker_information/*.parquet
+  - config_name: 8k_paralinguistic_information
+    data_files: data/8k/paralinguistic_information/*.parquet
+  - config_name: 8k_environmental_sound
+    data_files: data/8k/environmental_sound/*.parquet
+  - config_name: 16k_speech_semantics
+    data_files: data/16k/speech_semantics/*.parquet
+  - config_name: 16k_speaker_information
+    data_files: data/16k/speaker_information/*.parquet
+  - config_name: 16k_paralinguistic_information
+    data_files: data/16k/paralinguistic_information/*.parquet
+  - config_name: 16k_environmental_sound
+    data_files: data/16k/environmental_sound/*.parquet
+  - config_name: 32k_speech_semantics
+    data_files: data/32k/speech_semantics/*.parquet
+  - config_name: 32k_speaker_information
+    data_files: data/32k/speaker_information/*.parquet
+  - config_name: 32k_paralinguistic_information
+    data_files: data/32k/paralinguistic_information/*.parquet
+  - config_name: 32k_environmental_sound
+    data_files: data/32k/environmental_sound/*.parquet
+  - config_name: 64k_speech_semantics
+    data_files: data/64k/speech_semantics/*.parquet
+  - config_name: 64k_speaker_information
+    data_files: data/64k/speaker_information/*.parquet
+  - config_name: 64k_paralinguistic_information
+    data_files: data/64k/paralinguistic_information/*.parquet
+  - config_name: 64k_environmental_sound
+    data_files: data/64k/environmental_sound/*.parquet
+---
+
+# VoxMemBench: Long-Term Spoken Conversational Memory
+
+VoxMemBench evaluates whether a spoken-dialogue system remembers what it heard.
+A model is given many time-separated sessions of a conversation — every user
+turn as audio, every assistant turn as text — and is then asked a spoken
+question whose answer is somewhere in that history.
+
+**799 questions × 4 context lengths = 3,196 items.**
+
+What separates this from a text memory benchmark is *where the answer lives*.
+Three of the four evidence types are carried by the audio signal, not by the
+words, so a system that transcribes first and remembers later cannot reach them.
+
+| Evidence type | The answer depends on | Items |
+|---|---|---|
+| `speech_semantics` | what the user said | 928 |
+| `speaker_information` | who was speaking | 588 |
+| `paralinguistic_information` | how it was said — vocal delivery | 1,056 |
+| `environmental_sound` | what could be heard around the user | 624 |
+
+## Task
+
+```
+Session timestamp: 2025-06-17 09:27
+  user       <audio>
+  assistant  text
+  ...
+Session timestamp: 2025-06-23 01:52
+  user       <audio>
+  ...
+Session timestamp: 2025-09-01 07:44
+  user       <audio>      <- the question
+```
+
+Sessions are days or weeks apart. The model answers from the history it was
+given; it is never given a transcript.
+
+## Memory operations
+
+| Operation | The item asks the model to | Items |
+|---|---|---|
+| `information_extraction` | recover one fact from one session | 920 |
+| `multi_session_reasoning` | combine evidence across sessions | 884 |
+| `temporal_evolution_tracking` | track how something changed over time | 872 |
+| `answer_refusal` | recognise that the history does not contain the answer | 520 |
+
+Evidence type × memory operation gives **15 occupied cells**.
+`speech_semantics × information_extraction` is deliberately empty: reading one
+fact out of one transcript is not what this benchmark is for.
+
+## Context lengths
+
+Every question appears at four history lengths — 8K, 16K, 32K and 64K audio
+tokens — under one `question_id`. The question, the gold answer and the
+evidence are identical across the four; only the surrounding history grows, and
+it grows by nesting:
+
+```
+sessions(8K) ⊆ sessions(16K) ⊆ sessions(32K) ⊆ sessions(64K)
+```
+
+so a length comparison holds the item fixed and varies only the distraction.
+Four `latest_state` questions do not nest strictly — `q_00154`, `q_00158`,
+`q_00365`, `q_00645` — drop them from a strict length sweep.
+
+## What you download
+
+Every row carries its own history and every clip in it, so no config depends on
+any other. Sessions are shared between questions in the source data, but they
+are materialised into each item, which is what lets you take a slice and run it.
+
+The files are laid out by context length and then by evidence type:
+
+```
+data/<context length>/<evidence type>/train-000NN-of-000NN.parquet
+```
+
+and there is a config for each level, so you can take a whole context length,
+or a single evidence type within it:
+
+| Config | Items | 8K | 16K | 32K | 64K |
+|---|---|---|---|---|---|
+| `<length>` (all evidence) | 799 | 4.03 GB | 7.72 GB | 15.38 GB | 32.83 GB |
+| `<length>_speech_semantics` | 232 | 1.19 GB | 2.26 GB | 4.50 GB | 9.18 GB |
+| `<length>_speaker_information` | 147 | 0.73 GB | 1.41 GB | 2.80 GB | 6.52 GB |
+| `<length>_paralinguistic_information` | 264 | 1.33 GB | 2.55 GB | 5.12 GB | 10.38 GB |
+| `<length>_environmental_sound` | 156 | 0.79 GB | 1.50 GB | 2.96 GB | 6.75 GB |
+
+```python
+load_dataset("AudioMemory/voxmembench", "32k")                             # 799 items
+load_dataset("AudioMemory/voxmembench", "32k_paralinguistic_information")  # 264 items
+```
+
+Testing whether your system hears vocal delivery at 32K costs
+5.12 GB, not the whole benchmark.
+
+## History composition
+
+Every session in an item's history is labelled with its role:
+
+| Role | What it is |
+|---|---|
+| `evidence` | the session the answer comes from |
+| `samekey_haystack` | shares the question's retrieval key, and is ruled out only by the selector the question states |
+| `topical_haystack` | shares the topic but not the queried attribute |
+| `filler` | unrelated conversation |
+
+`samekey_haystack` is what stops length from being free. Without it, retrieval
+succeeds on topic words alone; with it, the model has to apply the selector the
+question actually states. It scales 1 / 2 / 4 / 8 with the context length. The
+16 answerable questions that carry none at any length — the question names only
+the topic, so nothing can be ruled out — are usable as a control group; filter
+on `samekey_haystack_session_count`. Refusal items carry no haystack by design.
+
+## Loading
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("AudioMemory/voxmembench", "32k", split="train")
+
+item = ds[0]
+print(item["question_text"], item["gold_json"])
+
+# the spoken question
+question = item["question_audio"]        # {"array": ..., "sampling_rate": 24000}
+
+# the history, in order, with every user clip already decoded
+for session in item["sessions"]:
+    print(session["timestamp"], session["role"])
+    for turn in session["turns"]:
+        if turn["audio"]:
+            audio = turn["audio"]["array"]
+        else:
+            assistant_text = turn["text"]
+```
+
+Audio arrives decoded because the shards declare their Hugging Face feature
+types; `datasets` needs its audio extra (`pip install "datasets[audio]"`). To
+avoid that dependency, read the parquet directly and decode the bytes yourself:
+
+```python
+import pyarrow.parquet as pq, soundfile as sf, io
+
+path = "data/32k/paralinguistic_information/train-00000-of-00005.parquet"
+row = pq.read_table(path).slice(0, 1).to_pylist()[0]
+wav, sr = sf.read(io.BytesIO(row["question_audio"]["bytes"]))
+```
+
+Streaming works too (`streaming=True`); row groups are about 128 MB, so a
+reader fetches one group rather than a whole shard to see a row.
+
+## Item schema
+
+One row per `(question_id, context_length)`, self-contained.
+
+| Field | Type | Description |
+|---|---|---|
+| `item_id` | string | `<question_id>_<context length>`, unique in the release |
+| `question_id` | string | stable across the four context lengths, e.g. `q_00142` |
+| `context_length` | string | `8K`, `16K`, `32K`, `64K` |
+| `family_id` | string | question family; one family may hold both an answerable and a refusal question |
+| `evidence_type` | string | one of the four above |
+| `memory_operation` | string | one of the four above |
+| `subtype` | string | finer question type, e.g. `cue_to_fact`, `latest_state`, `counting` |
+| `question_text` | string | transcript of the spoken question, for reference |
+| `question_audio` | audio | the spoken question |
+| `query_timestamp` | string | when the question is asked |
+| `gold_json` | string | the gold answer, JSON-encoded |
+| `answer_type` | string | `categorical`, `short_text`, `ordered_list`, `yes_no`, `number`, or null for refusal items |
+| `answer_normalization` | string | how to compare an answer to the gold |
+| `expected_response` | string | `answer`, or `insufficient_evidence` for refusal items |
+| `sessions` | list[session] | the history, in order |
+| `evidence_session_ids` | list[string] | the sessions holding the evidence |
+| `evidence_session_indices` | list[int] | their positions in `sessions` |
+| `evidence_relative_positions` | list[float] | the same positions as 0.0–1.0 |
+| `session_count`, `evidence_session_count`, `filler_session_count`, `samekey_haystack_session_count`, `topical_haystack_session_count` | int | history composition |
+| `history_audio_tokens`, `history_duration_seconds` | number | size of the history |
+| `paired_question_id` | string | the refusal counterpart of an answerable question, where one exists |
+
+A **session** has `session_id`, `timestamp`, `role` and `turns`. A **turn** has
+`role` (`user` or `assistant`), `text` and, on user turns, `audio`. All audio is
+24 kHz mono.
+
+A user turn carries **both** its audio and the transcript of it, so the
+benchmark can be read as well as heard — a text-only run is the ceiling the
+audio numbers are measured against. The transcript is the words only: the cue
+markers that drove the delivery are not in it, which is what keeps the
+paralinguistic, speaker and environmental items out of reach of a reader.
+
+`session_id` is stable across the whole release, so a session reused by two
+questions carries the same id in both — useful for caching an encoder's output,
+and for checking what a system has already seen.
+
+## Evaluation
+
+Give the model the system prompt, the ordered history as audio plus assistant
+text, and the final spoken question. Do not give it a transcript.
+
+```
+evaluation/candidate_system_prompt.txt             permits abstention
+evaluation/candidate_system_prompt.no_abstain.txt  always answer
+evaluation/answerable_judge_prompt.txt             grades answerable items
+evaluation/ar_judge_prompt.txt                     grades refusal items
+```
+
+**Metric.** Accuracy, reported separately for two strata:
+
+- *answerable* (2,676 items) — correct when the response is semantically
+  equivalent to `gold_json` under the item's `answer_normalization`. Answers are
+  open-ended, so equivalence is settled by an LLM judge that sees only the
+  question, the gold and the response — never the audio or the history.
+  `evaluation/answerable_judge_prompt.txt` is that contract. Abstaining is
+  incorrect.
+- *answer refusal* (520 items) — correct when the response says the evidence is
+  insufficient. Score this stratum only under the abstention-permitting prompt;
+  under the always-answer prompt it reads ~0 by construction.
+
+Report per context length, and per evidence type × memory operation when
+comparing systems: the aggregate hides that the audio-only cells behave nothing
+like the semantic ones.
+
+## How the data was made
+
+- **User speech** — voice cloned from 30 reference speakers of the
+  [CSTR VCTK Corpus](https://doi.org/10.7488/ds/2645) 0.92. The persona profiles
+  in `metadata/speakers.json` are fictional and are not recoverable from the
+  voice.
+- **Assistant turns** — text, never spoken.
+- **Environmental sound** — clips from
+  [ESC-50](https://doi.org/10.7910/DVN/YDEPUT), mixed into the user speech at a
+  target SNR rather than appended, so the sound is present *while* the user
+  talks.
+- **Paralinguistic cues** — rendered as vocal delivery in the cloned speech.
+
+## Public annotations
+
+Gold answers, evidence-session annotations and evidence positions are all
+public. There is no hidden split and no submission server: the benchmark is
+meant to be run locally and analysed — evidence-oracle conditions, retrieval
+error analysis, degradation against evidence position, modality ablations —
+which all need those fields.
+
+## Licence
+
+CC BY-NC 4.0. The non-commercial term follows from ESC-50, whose clips supply
+the environmental sound. Attribution is required for VCTK and for ESC-50; see
+`LICENSE`, and `metadata/environmental_sources.json` for the licence of every
+clip used.
+
+## Citation
+
+```bibtex
+@misc{voxmembench2026,
+  title  = {VoxMemBench: Long-Term Spoken Conversational Memory},
+  author = {The VoxMemBench authors},
+  year   = {2026},
+  url    = {https://huggingface.co/datasets/AudioMemory/voxmembench}
+}
+```

data/16k/environmental_sound/train-00000-of-00003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdf0fd2fdd23e8d3486ce0bd5a35a267d154f35a2653f3ed3f6c911a748ff596
+size 779474585

data/16k/environmental_sound/train-00001-of-00003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:acc34647f26ea936284cb60f2d1d2cc063fe3be32173e25325509796f6f75756
+size 741354732

data/16k/environmental_sound/train-00002-of-00003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07580ee7c136b6c9dbebd51c3f13888908f7a9b497e63731000c7a4f8770d927
+size 89371254

data/16k/paralinguistic_information/train-00000-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3c561a6cf2548283a827ee84ec8e6b701c8f9a8117c457be9e0a09b45810324
+size 775304879

data/16k/paralinguistic_information/train-00001-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca529cf1f8530a854d18eab759e9e35efd8fc4b16192c67010711d91782fe127
+size 816651653

data/16k/paralinguistic_information/train-00002-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c9a2656f2afa793fce15caf87266b2d389990e735c424594e94bee4a4bfa76d
+size 761863702

data/16k/paralinguistic_information/train-00003-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db84efd4aa0f7434356266b847bc7ae25615a5970dd713d39c7dad92b470e4a3
+size 386825926

data/16k/speaker_information/train-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:403da711afd0dc022aa2471734382f7b9738b772e8da2ecae7b19fee4aadc21f
+size 771937355

data/16k/speaker_information/train-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0888e915f5292a8cb998e1540c7d984094ce0d158578a0713b8cf3a8bf3cd0ae
+size 746626408

data/16k/speech_semantics/train-00000-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e29e4d8f61979ee397bdd2007b932abf81a17c8dd2b909a40633c3d735442c17
+size 811928940

data/16k/speech_semantics/train-00001-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:127106e1856996ac3781d85d9091db58bdb920e8823ca15c0bef73b300cf38ea
+size 761199259

data/16k/speech_semantics/train-00002-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd070c551357fe8ec5f400dd883b78a0f615628ed63539c0050a9306231e1b80
+size 771940345

data/16k/speech_semantics/train-00003-of-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a78906c3be7833cbf20a8210803d935c846770bc8b4a79d50946190507fc825f
+size 78620144

data/32k/environmental_sound/train-00000-of-00005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:379161e28757a19ceb99ba298ceb1571aa11bf045c11a230cf8d9ee2f3fdedee
+size 755108312

data/32k/environmental_sound/train-00001-of-00005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cadcc3149460607d5277297de032c75caeb6d001dd54cc37c0ecbd9083f6a4e
+size 827556304

data/32k/environmental_sound/train-00003-of-00005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6166c788fe74afb05ebb340745e27df51e837d21fa99dd2f1647ecdd6cd63f5
+size 726439675

data/64k/paralinguistic_information/train-00000-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:997ce4560638deffe253670d082ee6e32128324d8670b42c3a596fcf92dda006
+size 748113117

data/64k/paralinguistic_information/train-00001-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66bb8e0071d85a7098ef2b18b11df897d44dbeec4dd8c030a83db27b44022cf6
+size 828379635

data/64k/paralinguistic_information/train-00002-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e8b0bb5869cbea55f70b59bde26a60e4eb204bc0ee4096d7a4285f3c196ba7f
+size 741918331

data/64k/paralinguistic_information/train-00003-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c75e0b570df7a055d3ae07c8dc8863bfc3939eb985a1b8233d43ec131ca8819c
+size 803911640

data/64k/paralinguistic_information/train-00004-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9814d7a29629d40374f07c70939eddb3ce038abfa73ec65c3507431ff00aebf
+size 876433089

data/64k/paralinguistic_information/train-00005-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f171a3d0ac130acb4db9d59f2baa3eb8a655eb306444f19df269ad0ae404be6
+size 854484447

data/64k/paralinguistic_information/train-00006-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91b45acfae6e6110b363e943c16029d83e6a023fd8d422636f53b19fd7ca0346
+size 851651101

data/64k/paralinguistic_information/train-00007-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee47a9de5a8a84c32eb3d904c8829802189e9dda7107bbf4ac5c890ef10f1761
+size 805643401

data/64k/paralinguistic_information/train-00008-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e61ff6e97d46e82ca22831fc7f6e365d71430784b8d503da1d2faa33645caf7a
+size 689396853

data/64k/paralinguistic_information/train-00009-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3e00c889bb9bac8b25d62a4413593f8a3ffbe08db82e6ae36fd7ddaef55bdea
+size 880046968

data/64k/paralinguistic_information/train-00010-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e23594791c293ae4ffc903b6b4798e636d58eaf7821315b4302726be3e6463d1
+size 827258851

data/64k/paralinguistic_information/train-00011-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccb2286918dbad5bda0b72f1d770142d547e265531ce5862ba7f68b4b097a74c
+size 811479027

data/64k/paralinguistic_information/train-00012-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81c045d296a9e28d74d94930b581c81c29348842bf551f2718e00e6f21278606
+size 751733813

data/64k/paralinguistic_information/train-00013-of-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f4a60e68a91a4f4d36b69f70e974e5bc63b45c430bb88eca2c7d939f0741631
+size 676135328

data/64k/speech_semantics/train-00009-of-00013.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34b753c81f3691939397dfb00c55a3022f64be2896ed040e459ed2d43f95169c
+size 804450681

data/8k/environmental_sound/train-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3609276a89c11b7bd9aa9d331a354390a51cfdb9e61786132284faf4cda74b71
+size 771080187

data/8k/environmental_sound/train-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e05db715de709dd1da7a61ae43f067d22db220fb44744c7e0d629dc8dc591dbf
+size 75470649

data/8k/paralinguistic_information/train-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b17864ccde544cd0369147e15e2f65f572b2c104e475feb324ff4660e59c1d0
+size 787021204

data/8k/paralinguistic_information/train-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5cbc23342424e9877c78bf4a48bb2796d3cb6c924a6a8e92380d40cacaaabd4
+size 639886033

data/8k/speaker_information/train-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1949bae0b07c390227c11c0ba75dc852de9ae9177190dd49120284562f0e4f03
+size 772957460

data/8k/speaker_information/train-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d96d24f69c86d5df02ee4dd094c094e9ee1562e158d3005c98ccfeb1ce593427
+size 5803626

data/8k/speech_semantics/train-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2711c5d99f7b218e2b123001d9693de344e7d27f7b88515d368c8677365c56fd
+size 789224557

data/8k/speech_semantics/train-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81aa6170b4a7955fdf9bc276105d924b63273455557f188ea911647fa286aed8
+size 484128743

evaluation/answerable_judge_prompt.txt

@@ -0,0 +1,37 @@
+You are evaluating the correctness of a model's answer to a conversational memory question.
+
+You are given:
+- the final question,
+- the reference answer,
+- the expected answer type,
+- and the model's response.
+
+Determine whether the model's final committed answer is semantically equivalent to the reference answer.
+
+Evaluation rules:
+
+1. Judge correctness by meaning, not exact wording.
+2. Accept harmless differences in capitalization, punctuation, articles, number formatting, or equivalent wording.
+3. For numerical answers, accept numerically equivalent expressions.
+4. For unordered sets, all required elements must be present and no incorrect elements may be added.
+5. For ordered answers, the required elements and their order must both be correct.
+6. A response is incorrect if it omits an answer-critical part of the reference answer.
+7. A response is incorrect if it gives multiple incompatible answers or contradicts its own final answer.
+8. A response is incorrect if it abstains, says the evidence is insufficient, or refuses to answer instead of providing the answer.
+9. Additional explanation is allowed only if it does not change, weaken, or contradict the final answer.
+10. Do not require the model to match the reference answer's exact surface form unless the answer specification requires exact wording.
+
+When an ANSWER OPTIONS list is given, these rules apply in addition:
+
+11. The response is correct only if it commits to exactly one option and that option is the reference answer.
+12. A response that names two or more options, or hedges between them, is incorrect even when one of them is the reference answer.
+13. A response that does not quote an option verbatim is resolved by meaning: if it is semantically equivalent to exactly one option, treat it as that option; if it matches none of them, or more than one, it is incorrect.
+14. For an ordered option, every element and their order must match; a response that names the right sounds in the wrong order is incorrect.
+
+Return only:
+
+{"verdict":"correct"}
+
+or
+
+{"verdict":"incorrect"}

evaluation/ar_judge_prompt.txt

@@ -0,0 +1,22 @@
+You are evaluating whether a model correctly abstained from answering an unanswerable conversational memory question.
+
+The correct behavior is to indicate that the provided conversation does not contain enough evidence to determine one unique answer.
+
+Evaluate the model response using the following rules:
+
+1. A response is correct if it clearly states that the answer cannot be determined because the available conversational evidence is insufficient, missing, incomplete, or ambiguous.
+2. The exact phrase "insufficient evidence" is not required. Semantically equivalent evidence-based abstentions are acceptable.
+3. A generic statement such as "I don't know" is not sufficient unless it clearly attributes the uncertainty to missing or insufficient information in the provided conversation.
+4. A safety, policy, capability, or willingness-based refusal does not count as a correct abstention.
+5. A response is incorrect if it provides or commits to a specific answer.
+6. A response is incorrect if it first states that the evidence is insufficient but then guesses or proposes a likely answer.
+7. A response is incorrect if it presents several candidate answers instead of abstaining.
+8. Do not judge whether the question itself is actually answerable. Assume the benchmark annotation is correct and evaluate only whether the model response exhibits the required evidence-based abstention behavior.
+
+Return only:
+
+{"verdict":"correct"}
+
+or
+
+{"verdict":"incorrect"}

evaluation/candidate_system_prompt.no_abstain.txt

@@ -0,0 +1,7 @@
+You are continuing an ongoing conversation that took place across multiple timestamped sessions.
+
+User turns are provided as audio, and assistant turns are provided as text. The final user audio is the current query.
+
+Use the complete conversation history to answer the final query. Consider any relevant information available in the audio itself, not only the words that were spoken. Different user audio turns may or may not be spoken by the same person.
+
+Base your answer only on the provided conversation. Always give your best answer, even if you are not certain. Provide a concise answer without explaining your reasoning.

evaluation/candidate_system_prompt.txt

@@ -0,0 +1,11 @@
+You are continuing an ongoing conversation that took place across multiple timestamped sessions.
+
+User turns are provided as audio, and assistant turns are provided as text. The final user audio is the current query.
+
+Use the complete conversation history to answer the final query. Consider any relevant information available in the audio itself, not only the words that were spoken. Different user audio turns may or may not be spoken by the same person.
+
+Base your answer only on the provided conversation. If the conversation does not contain enough information to determine one unique answer, respond exactly with:
+
+insufficient evidence
+
+Otherwise, provide a concise answer without explaining your reasoning.

metadata/environmental_sources.json

@@ -0,0 +1,325 @@
+{
+  "clips": [
+    {
+      "asset_id": "ESC50_1-100032-A-0",
+      "category": "dog",
+      "clip_license": "CC0",
+      "esc50_clip": "1-100032-A-0.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-100038-A-14",
+      "category": "chirping_birds",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-100038-A-14.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-100210-A-36",
+      "category": "vacuum_cleaner",
+      "clip_license": "CC-BY-NC",
+      "esc50_clip": "1-100210-A-36.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-101296-A-19",
+      "category": "thunderstorm",
+      "clip_license": "CC0",
+      "esc50_clip": "1-101296-A-19.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-101336-A-30",
+      "category": "door_wood_knock",
+      "clip_license": "CC-BY-NC",
+      "esc50_clip": "1-101336-A-30.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-101404-A-34",
+      "category": "can_opening",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-101404-A-34.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-103298-A-9",
+      "category": "crow",
+      "clip_license": "CC-Sampling+",
+      "esc50_clip": "1-103298-A-9.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-104089-A-22",
+      "category": "clapping",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-104089-A-22.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-115545-A-48",
+      "category": "fireworks",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-115545-A-48.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-116765-A-41",
+      "category": "chainsaw",
+      "clip_license": "CC0",
+      "esc50_clip": "1-116765-A-41.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-11687-A-47",
+      "category": "airplane",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-11687-A-47.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-118206-A-31",
+      "category": "mouse_click",
+      "clip_license": "CC0",
+      "esc50_clip": "1-118206-A-31.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-118559-A-17",
+      "category": "pouring_water",
+      "clip_license": "CC0",
+      "esc50_clip": "1-118559-A-17.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-119125-A-45",
+      "category": "train",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-119125-A-45.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-121951-A-8",
+      "category": "sheep",
+      "clip_license": "CC0",
+      "esc50_clip": "1-121951-A-8.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-12653-A-15",
+      "category": "water_drops",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-12653-A-15.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-13571-A-46",
+      "category": "church_bells",
+      "clip_license": "CC-BY-NC",
+      "esc50_clip": "1-13571-A-46.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-13613-A-37",
+      "category": "clock_alarm",
+      "clip_license": "CC0",
+      "esc50_clip": "1-13613-A-37.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-137-A-32",
+      "category": "keyboard_typing",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-137-A-32.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-137296-A-16",
+      "category": "wind",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-137296-A-16.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-155858-A-25",
+      "category": "footsteps",
+      "clip_license": "CC0",
+      "esc50_clip": "1-155858-A-25.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-15689-A-4",
+      "category": "frog",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-15689-A-4.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-16568-A-3",
+      "category": "cow",
+      "clip_license": "CC-BY-NC",
+      "esc50_clip": "1-16568-A-3.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-17092-A-27",
+      "category": "brushing_teeth",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-17092-A-27.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-17124-A-43",
+      "category": "car_horn",
+      "clip_license": "CC0",
+      "esc50_clip": "1-17124-A-43.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-17150-A-12",
+      "category": "crackling_fire",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-17150-A-12.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-172649-A-40",
+      "category": "helicopter",
+      "clip_license": "CC0",
+      "esc50_clip": "1-172649-A-40.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-17295-A-29",
+      "category": "drinking_sipping",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-17295-A-29.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-17367-A-10",
+      "category": "rain",
+      "clip_license": "CC0",
+      "esc50_clip": "1-17367-A-10.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-17585-A-7",
+      "category": "insects",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-17585-A-7.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-18074-A-6",
+      "category": "hen",
+      "clip_license": "CC-BY-NC",
+      "esc50_clip": "1-18074-A-6.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-18527-A-44",
+      "category": "engine",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-18527-A-44.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-187207-A-20",
+      "category": "crying_baby",
+      "clip_license": "CC0",
+      "esc50_clip": "1-187207-A-20.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-18810-A-49",
+      "category": "hand_saw",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-18810-A-49.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-20133-A-39",
+      "category": "glass_breaking",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-20133-A-39.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-20736-A-18",
+      "category": "toilet_flush",
+      "clip_license": "CC-Sampling+",
+      "esc50_clip": "1-20736-A-18.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-208757-A-2",
+      "category": "pig",
+      "clip_license": "CC0",
+      "esc50_clip": "1-208757-A-2.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-21896-A-35",
+      "category": "washing_machine",
+      "clip_license": "CC-Sampling+",
+      "esc50_clip": "1-21896-A-35.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-21934-A-38",
+      "category": "clock_tick",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-21934-A-38.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-26806-A-1",
+      "category": "rooster",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-26806-A-1.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-28135-A-11",
+      "category": "sea_waves",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-28135-A-11.wav",
+      "in_esc10_subset": true
+    },
+    {
+      "asset_id": "ESC50_1-31482-A-42",
+      "category": "siren",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-31482-A-42.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-34094-A-5",
+      "category": "cat",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-34094-A-5.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-51805-A-33",
+      "category": "door_wood_creaks",
+      "clip_license": "CC0",
+      "esc50_clip": "1-51805-A-33.wav",
+      "in_esc10_subset": false
+    },
+    {
+      "asset_id": "ESC50_1-57316-A-13",
+      "category": "crickets",
+      "clip_license": "CC-BY",
+      "esc50_clip": "1-57316-A-13.wav",
+      "in_esc10_subset": false
+    }
+  ],
+  "source_dataset": {
+    "dataset_license": "CC BY-NC 3.0 (the ESC-10 subset is CC BY 3.0)",
+    "dataset_name": "ESC-50",
+    "dataset_ref": "doi:10.7910/DVN/YDEPUT",
+    "note": "Individual clips carry their own Freesound licences, listed per clip below. Clips that are not CC0 or CC-BY restrict how the sessions mixed from them may be reused."
+  }
+}

metadata/speakers.json

@@ -0,0 +1,342 @@
+{
+  "note": "Every user voice in the benchmark is a clone conditioned on one VCTK reference speaker. The persona profile is fictional and is not recoverable from the voice.",
+  "source_dataset": {
+    "dataset_name": "CSTR VCTK Corpus",
+    "dataset_ref": "doi:10.7488/ds/2645",
+    "dataset_version": "0.92",
+    "language": "en",
+    "license_id": "CC-BY-4.0"
+  },
+  "speakers": [
+    {
+      "persona_age": 24,
+      "persona_age_tier": "young adult",
+      "persona_gender": "female",
+      "source_age": 24,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p283",
+      "speaker_id": "spk_001"
+    },
+    {
+      "persona_age": 27,
+      "persona_age_tier": "young adult",
+      "persona_gender": "male",
+      "source_age": 23,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p273",
+      "speaker_id": "spk_002"
+    },
+    {
+      "persona_age": 22,
+      "persona_age_tier": "young adult",
+      "persona_gender": "female",
+      "source_age": 22,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p228",
+      "speaker_id": "spk_003"
+    },
+    {
+      "persona_age": 29,
+      "persona_age_tier": "young adult",
+      "persona_gender": "male",
+      "source_age": 23,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p272",
+      "speaker_id": "spk_004"
+    },
+    {
+      "persona_age": 25,
+      "persona_age_tier": "young adult",
+      "persona_gender": "female",
+      "source_age": 24,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p276",
+      "speaker_id": "spk_005"
+    },
+    {
+      "persona_age": 23,
+      "persona_age_tier": "young adult",
+      "persona_gender": "male",
+      "source_age": 23,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p279",
+      "speaker_id": "spk_006"
+    },
+    {
+      "persona_age": 28,
+      "persona_age_tier": "young adult",
+      "persona_gender": "female",
+      "source_age": 24,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p257",
+      "speaker_id": "spk_007"
+    },
+    {
+      "persona_age": 26,
+      "persona_age_tier": "young adult",
+      "persona_gender": "male",
+      "source_age": 23,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p275",
+      "speaker_id": "spk_008"
+    },
+    {
+      "persona_age": 30,
+      "persona_age_tier": "young adult",
+      "persona_gender": "female",
+      "source_age": 25,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p335",
+      "speaker_id": "spk_009"
+    },
+    {
+      "persona_age": 21,
+      "persona_age_tier": "young adult",
+      "persona_gender": "male",
+      "source_age": 21,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p241",
+      "speaker_id": "spk_010"
+    },
+    {
+      "persona_age": 35,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "female",
+      "source_age": 25,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p299",
+      "speaker_id": "spk_011"
+    },
+    {
+      "persona_age": 42,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "male",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p347",
+      "speaker_id": "spk_012"
+    },
+    {
+      "persona_age": 38,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "female",
+      "source_age": 25,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p280",
+      "speaker_id": "spk_013"
+    },
+    {
+      "persona_age": 33,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "male",
+      "source_age": 23,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p232",
+      "speaker_id": "spk_014"
+    },
+    {
+      "persona_age": 40,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "female",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p330",
+      "speaker_id": "spk_015"
+    },
+    {
+      "persona_age": 31,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "male",
+      "source_age": 23,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p259",
+      "speaker_id": "spk_016"
+    },
+    {
+      "persona_age": 44,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "female",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p314",
+      "speaker_id": "spk_017"
+    },
+    {
+      "persona_age": 36,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "male",
+      "source_age": 25,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p245",
+      "speaker_id": "spk_018"
+    },
+    {
+      "persona_age": 39,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "female",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p341",
+      "speaker_id": "spk_019"
+    },
+    {
+      "persona_age": 34,
+      "persona_age_tier": "mid-age",
+      "persona_gender": "male",
+      "source_age": 24,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p256",
+      "speaker_id": "spk_020"
+    },
+    {
+      "persona_age": 52,
+      "persona_age_tier": "mature",
+      "persona_gender": "male",
+      "source_age": 28,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p374",
+      "speaker_id": "spk_021"
+    },
+    {
+      "persona_age": 48,
+      "persona_age_tier": "mature",
+      "persona_gender": "female",
+      "source_age": 27,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p343",
+      "speaker_id": "spk_022"
+    },
+    {
+      "persona_age": 55,
+      "persona_age_tier": "mature",
+      "persona_gender": "male",
+      "source_age": 29,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p281",
+      "speaker_id": "spk_023"
+    },
+    {
+      "persona_age": 46,
+      "persona_age_tier": "mature",
+      "persona_gender": "female",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p261",
+      "speaker_id": "spk_024"
+    },
+    {
+      "persona_age": 58,
+      "persona_age_tier": "mature",
+      "persona_gender": "male",
+      "source_age": 38,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p227",
+      "speaker_id": "spk_025"
+    },
+    {
+      "persona_age": 50,
+      "persona_age_tier": "mature",
+      "persona_gender": "female",
+      "source_age": 29,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p362",
+      "speaker_id": "spk_026"
+    },
+    {
+      "persona_age": 47,
+      "persona_age_tier": "mature",
+      "persona_gender": "male",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p326",
+      "speaker_id": "spk_027"
+    },
+    {
+      "persona_age": 54,
+      "persona_age_tier": "mature",
+      "persona_gender": "female",
+      "source_age": 32,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p318",
+      "speaker_id": "spk_028"
+    },
+    {
+      "persona_age": 49,
+      "persona_age_tier": "mature",
+      "persona_gender": "male",
+      "source_age": 26,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "male",
+      "source_speaker_id": "p251",
+      "speaker_id": "spk_029"
+    },
+    {
+      "persona_age": 57,
+      "persona_age_tier": "mature",
+      "persona_gender": "female",
+      "source_age": 33,
+      "source_dataset": "CSTR VCTK Corpus",
+      "source_dataset_version": "0.92",
+      "source_gender": "female",
+      "source_speaker_id": "p294",
+      "speaker_id": "spk_030"
+    }
+  ]
+}

metadata/statistics.json

@@ -0,0 +1,80 @@
+{
+  "benchmark_version": "1.0.0",
+  "counts_by_context_length": {
+    "16K": 799,
+    "32K": 799,
+    "64K": 799,
+    "8K": 799
+  },
+  "counts_by_evidence_type": {
+    "environmental_sound": 624,
+    "paralinguistic_information": 1056,
+    "speaker_information": 588,
+    "speech_semantics": 928
+  },
+  "counts_by_evidence_type_and_memory_operation": {
+    "environmental_sound|answer_refusal": 120,
+    "environmental_sound|information_extraction": 232,
+    "environmental_sound|multi_session_reasoning": 144,
+    "environmental_sound|temporal_evolution_tracking": 128,
+    "paralinguistic_information|answer_refusal": 168,
+    "paralinguistic_information|information_extraction": 428,
+    "paralinguistic_information|multi_session_reasoning": 264,
+    "paralinguistic_information|temporal_evolution_tracking": 196,
+    "speaker_information|answer_refusal": 84,
+    "speaker_information|information_extraction": 260,
+    "speaker_information|multi_session_reasoning": 136,
+    "speaker_information|temporal_evolution_tracking": 108,
+    "speech_semantics|answer_refusal": 148,
+    "speech_semantics|multi_session_reasoning": 340,
+    "speech_semantics|temporal_evolution_tracking": 440
+  },
+  "counts_by_memory_operation": {
+    "answer_refusal": 520,
+    "information_extraction": 920,
+    "multi_session_reasoning": 884,
+    "temporal_evolution_tracking": 872
+  },
+  "num_distinct_audio_clips": 154697,
+  "num_distinct_sessions": 34743,
+  "num_families": 669,
+  "num_items": 3196,
+  "num_questions": 799,
+  "parquet_shards": {
+    "16k/environmental_sound": 3,
+    "16k/paralinguistic_information": 4,
+    "16k/speaker_information": 2,
+    "16k/speech_semantics": 4,
+    "32k/environmental_sound": 5,
+    "32k/paralinguistic_information": 7,
+    "32k/speaker_information": 4,
+    "32k/speech_semantics": 7,
+    "64k/environmental_sound": 10,
+    "64k/paralinguistic_information": 14,
+    "64k/speaker_information": 9,
+    "64k/speech_semantics": 13,
+    "8k/environmental_sound": 2,
+    "8k/paralinguistic_information": 2,
+    "8k/speaker_information": 2,
+    "8k/speech_semantics": 2
+  },
+  "released_size_bytes": 64394412457,
+  "size_bytes_by_config": {
+    "16k/environmental_sound": 1610200571,
+    "16k/paralinguistic_information": 2740646160,
+    "16k/speaker_information": 1518563763,
+    "16k/speech_semantics": 2423688688,
+    "32k/environmental_sound": 3183302286,
+    "32k/paralinguistic_information": 5499951257,
+    "32k/speaker_information": 3002392450,
+    "32k/speech_semantics": 4829808146,
+    "64k/environmental_sound": 7253896286,
+    "64k/paralinguistic_information": 11146585601,
+    "64k/speaker_information": 7000138465,
+    "64k/speech_semantics": 9859666325,
+    "8k/environmental_sound": 846550836,
+    "8k/paralinguistic_information": 1426907237,
+    "8k/speaker_information": 778761086,
+    "8k/speech_semantics": 1273353300
+  }
+}