Spaces:
Running on Zero
Running on Zero
Vansh Chugh commited on
Commit ·
db50e80
1
Parent(s): 6017d6d
fail soft on unmatched instrument names, add matching-status output
Browse files
app.py
CHANGED
|
@@ -18,13 +18,14 @@ except ImportError:
|
|
| 18 |
return func
|
| 19 |
|
| 20 |
import tempfile
|
|
|
|
| 21 |
|
| 22 |
import gradio as gr
|
| 23 |
import torch
|
| 24 |
from pyharp import ModelCard, build_endpoint
|
| 25 |
|
| 26 |
from muscriptor import TranscriptionModel
|
| 27 |
-
from muscriptor.tokenizer.mt3 import resolve_instrument_names
|
| 28 |
|
| 29 |
model_card = ModelCard(
|
| 30 |
name="MuScriptor",
|
|
@@ -36,6 +37,8 @@ model_card = ModelCard(
|
|
| 36 |
tags=["transcription", "midi", "multi-instrument"],
|
| 37 |
)
|
| 38 |
|
|
|
|
|
|
|
| 39 |
_models: dict[str, TranscriptionModel] = {}
|
| 40 |
|
| 41 |
|
|
@@ -51,19 +54,34 @@ def _get_model(variant: str) -> TranscriptionModel:
|
|
| 51 |
return _models[variant]
|
| 52 |
|
| 53 |
|
| 54 |
-
def
|
| 55 |
-
"""
|
| 56 |
|
| 57 |
-
|
| 58 |
-
|
|
|
|
| 59 |
"""
|
| 60 |
tokens = [t for t in text.split(",") if t.strip()]
|
| 61 |
if not tokens:
|
| 62 |
-
return None
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
| 68 |
|
| 69 |
@spaces.GPU
|
|
@@ -74,10 +92,10 @@ def process_fn(
|
|
| 74 |
instruments_text: str,
|
| 75 |
use_sampling: bool,
|
| 76 |
temperature: float,
|
| 77 |
-
) -> str:
|
| 78 |
-
"""Transcribe the input audio to
|
| 79 |
model = _get_model(variant)
|
| 80 |
-
instruments =
|
| 81 |
|
| 82 |
midi_bytes = model.transcribe_to_midi(
|
| 83 |
input_audio_path,
|
|
@@ -90,7 +108,12 @@ def process_fn(
|
|
| 90 |
f.write(midi_bytes)
|
| 91 |
output_midi_path = f.name
|
| 92 |
|
| 93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
|
| 95 |
|
| 96 |
with gr.Blocks() as demo:
|
|
@@ -105,7 +128,7 @@ with gr.Blocks() as demo:
|
|
| 105 |
gr.Textbox(
|
| 106 |
value="",
|
| 107 |
label="Instruments",
|
| 108 |
-
info="Comma-separated instrument names to restrict decoding to
|
| 109 |
),
|
| 110 |
gr.Checkbox(
|
| 111 |
label="Use Sampling",
|
|
@@ -124,6 +147,9 @@ with gr.Blocks() as demo:
|
|
| 124 |
gr.File(type="filepath", file_types=[".mid", ".midi"], label="Output MIDI").set_info(
|
| 125 |
"Transcribed MIDI notes."
|
| 126 |
),
|
|
|
|
|
|
|
|
|
|
| 127 |
]
|
| 128 |
|
| 129 |
build_endpoint(
|
|
|
|
| 18 |
return func
|
| 19 |
|
| 20 |
import tempfile
|
| 21 |
+
from pathlib import Path
|
| 22 |
|
| 23 |
import gradio as gr
|
| 24 |
import torch
|
| 25 |
from pyharp import ModelCard, build_endpoint
|
| 26 |
|
| 27 |
from muscriptor import TranscriptionModel
|
| 28 |
+
from muscriptor.tokenizer.mt3 import MT3_FULL_PLUS_GROUP_NAMES, resolve_instrument_names
|
| 29 |
|
| 30 |
model_card = ModelCard(
|
| 31 |
name="MuScriptor",
|
|
|
|
| 37 |
tags=["transcription", "midi", "multi-instrument"],
|
| 38 |
)
|
| 39 |
|
| 40 |
+
_VALID_INSTRUMENTS = ", ".join(MT3_FULL_PLUS_GROUP_NAMES)
|
| 41 |
+
|
| 42 |
_models: dict[str, TranscriptionModel] = {}
|
| 43 |
|
| 44 |
|
|
|
|
| 54 |
return _models[variant]
|
| 55 |
|
| 56 |
|
| 57 |
+
def _resolve_instruments(text: str) -> tuple[list[str] | None, str]:
|
| 58 |
+
"""Resolve the comma-separated instruments box into exact group names.
|
| 59 |
|
| 60 |
+
Names that don't resolve are dropped rather than raised: HARP's client
|
| 61 |
+
shows generic error message, so an error here would not reach the user.
|
| 62 |
+
Second value returns what happened as a .txt output
|
| 63 |
"""
|
| 64 |
tokens = [t for t in text.split(",") if t.strip()]
|
| 65 |
if not tokens:
|
| 66 |
+
return None, "No instrument restriction requested."
|
| 67 |
+
|
| 68 |
+
resolved: list[str] = []
|
| 69 |
+
problems: list[str] = []
|
| 70 |
+
for token in tokens:
|
| 71 |
+
try:
|
| 72 |
+
resolved.extend(resolve_instrument_names([token]))
|
| 73 |
+
except ValueError as e:
|
| 74 |
+
problems.append(str(e))
|
| 75 |
+
|
| 76 |
+
if not problems:
|
| 77 |
+
return resolved, f"Matched: {', '.join(resolved)}."
|
| 78 |
+
|
| 79 |
+
note = "; ".join(problems)
|
| 80 |
+
if resolved:
|
| 81 |
+
note = f"Matched: {', '.join(resolved)}. Ignored: {note}"
|
| 82 |
+
else:
|
| 83 |
+
note = f"No valid instrument names found, transcribing unrestricted. {note}"
|
| 84 |
+
return (resolved or None), note
|
| 85 |
|
| 86 |
|
| 87 |
@spaces.GPU
|
|
|
|
| 92 |
instruments_text: str,
|
| 93 |
use_sampling: bool,
|
| 94 |
temperature: float,
|
| 95 |
+
) -> tuple[str, str]:
|
| 96 |
+
"""Transcribe the input audio to MIDI, plus a note on the Instruments field."""
|
| 97 |
model = _get_model(variant)
|
| 98 |
+
instruments, instrument_note = _resolve_instruments(instruments_text)
|
| 99 |
|
| 100 |
midi_bytes = model.transcribe_to_midi(
|
| 101 |
input_audio_path,
|
|
|
|
| 108 |
f.write(midi_bytes)
|
| 109 |
output_midi_path = f.name
|
| 110 |
|
| 111 |
+
notes_text = f"{Path(input_audio_path).name}\n\nInstrument Matching\n{instrument_note}\n"
|
| 112 |
+
with tempfile.NamedTemporaryFile(mode="w", suffix=".txt", delete=False) as f:
|
| 113 |
+
f.write(notes_text)
|
| 114 |
+
output_notes_path = f.name
|
| 115 |
+
|
| 116 |
+
return output_midi_path, output_notes_path
|
| 117 |
|
| 118 |
|
| 119 |
with gr.Blocks() as demo:
|
|
|
|
| 128 |
gr.Textbox(
|
| 129 |
value="",
|
| 130 |
label="Instruments",
|
| 131 |
+
info=f"Comma-separated instrument names to restrict decoding to. Valid names: {_VALID_INSTRUMENTS}. Leave blank to let the model detect instruments on its own.",
|
| 132 |
),
|
| 133 |
gr.Checkbox(
|
| 134 |
label="Use Sampling",
|
|
|
|
| 147 |
gr.File(type="filepath", file_types=[".mid", ".midi"], label="Output MIDI").set_info(
|
| 148 |
"Transcribed MIDI notes."
|
| 149 |
),
|
| 150 |
+
gr.File(type="filepath", file_types=[".txt"], label="Instrument Matching").set_info(
|
| 151 |
+
"Which requested instrument names were matched or ignored."
|
| 152 |
+
),
|
| 153 |
]
|
| 154 |
|
| 155 |
build_endpoint(
|