Spaces:
Running
on
Zero
Running
on
Zero
update to v1.1
Browse files
app.py
CHANGED
@@ -7,21 +7,16 @@ from typing import Optional, List, Dict, Any
|
|
7 |
import torch
|
8 |
import gradio as gr
|
9 |
import yt_dlp as youtube_dl
|
10 |
-
import numpy as np
|
11 |
from transformers import pipeline
|
12 |
from transformers.pipelines.audio_utils import ffmpeg_read
|
13 |
-
from punctuators.models import PunctCapSegModelONNX
|
14 |
-
from stable_whisper import WhisperResult
|
15 |
|
16 |
|
17 |
# configuration
|
18 |
-
MODEL_NAME = "kotoba-tech/kotoba-whisper-v1.
|
19 |
BATCH_SIZE = 16
|
20 |
CHUNK_LENGTH_S = 15
|
21 |
FILE_LIMIT_MB = 1000
|
22 |
YT_LENGTH_LIMIT_S = 3600 # limit to 1 hour YouTube files
|
23 |
-
|
24 |
-
|
25 |
# device setting
|
26 |
if torch.cuda.is_available():
|
27 |
torch_dtype = torch.bfloat16
|
@@ -31,117 +26,19 @@ else:
|
|
31 |
torch_dtype = torch.float32
|
32 |
device = "cpu"
|
33 |
model_kwargs = {}
|
34 |
-
|
35 |
# define the pipeline
|
36 |
pipe = pipeline(
|
37 |
-
task="automatic-speech-recognition",
|
38 |
model=MODEL_NAME,
|
39 |
chunk_length_s=CHUNK_LENGTH_S,
|
40 |
batch_size=BATCH_SIZE,
|
41 |
torch_dtype=torch_dtype,
|
42 |
device=device,
|
43 |
-
model_kwargs=model_kwargs
|
|
|
|
|
44 |
)
|
45 |
|
46 |
|
47 |
-
class Punctuator:
|
48 |
-
|
49 |
-
ja_punctuations = ["!", "?", "γ", "γ"]
|
50 |
-
|
51 |
-
def __init__(self, model: str = "pcs_47lang"):
|
52 |
-
self.punctuation_model = PunctCapSegModelONNX.from_pretrained(model)
|
53 |
-
|
54 |
-
def punctuate(self, pipeline_chunk: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
|
55 |
-
|
56 |
-
def validate_punctuation(raw: str, punctuated: str):
|
57 |
-
if 'unk' in punctuated:
|
58 |
-
return raw
|
59 |
-
if punctuated.count("γ") > 1:
|
60 |
-
ind = punctuated.rfind("γ")
|
61 |
-
punctuated = punctuated.replace("γ", "")
|
62 |
-
punctuated = punctuated[:ind] + "γ" + punctuated[ind:]
|
63 |
-
return punctuated
|
64 |
-
|
65 |
-
text_edit = self.punctuation_model.infer([c['text'] for c in pipeline_chunk])
|
66 |
-
return [
|
67 |
-
{
|
68 |
-
'timestamp': c['timestamp'],
|
69 |
-
'text': validate_punctuation(c['text'], "".join(e))
|
70 |
-
} for c, e in zip(pipeline_chunk, text_edit)
|
71 |
-
]
|
72 |
-
|
73 |
-
|
74 |
-
PUNCTUATOR = Punctuator()
|
75 |
-
|
76 |
-
|
77 |
-
def _fix_timestamp(sample_rate: int, result: List[Dict[str, Any]], audio: np.ndarray) -> WhisperResult or None:
|
78 |
-
|
79 |
-
def replace_none_ts(parts):
|
80 |
-
total_dur = round(audio.shape[-1] / sample_rate, 3)
|
81 |
-
_medium_dur = _ts_nonzero_mask = None
|
82 |
-
|
83 |
-
def ts_nonzero_mask() -> np.ndarray:
|
84 |
-
nonlocal _ts_nonzero_mask
|
85 |
-
if _ts_nonzero_mask is None:
|
86 |
-
_ts_nonzero_mask = np.array([(p['end'] or p['start']) is not None for p in parts])
|
87 |
-
return _ts_nonzero_mask
|
88 |
-
|
89 |
-
def medium_dur() -> float:
|
90 |
-
nonlocal _medium_dur
|
91 |
-
if _medium_dur is None:
|
92 |
-
nonzero_dus = [p['end'] - p['start'] for p in parts if None not in (p['end'], p['start'])]
|
93 |
-
nonzero_durs = np.array(nonzero_dus)
|
94 |
-
_medium_dur = np.median(nonzero_durs) * 2 if len(nonzero_durs) else 2.0
|
95 |
-
return _medium_dur
|
96 |
-
|
97 |
-
def _curr_max_end(start: float, next_idx: float) -> float:
|
98 |
-
max_end = total_dur
|
99 |
-
if next_idx != len(parts):
|
100 |
-
mask = np.flatnonzero(ts_nonzero_mask()[next_idx:])
|
101 |
-
if len(mask):
|
102 |
-
_part = parts[mask[0]+next_idx]
|
103 |
-
max_end = _part['start'] or _part['end']
|
104 |
-
|
105 |
-
new_end = round(start + medium_dur(), 3)
|
106 |
-
if new_end > max_end:
|
107 |
-
return max_end
|
108 |
-
return new_end
|
109 |
-
|
110 |
-
for i, part in enumerate(parts, 1):
|
111 |
-
if part['start'] is None:
|
112 |
-
is_first = i == 1
|
113 |
-
if is_first:
|
114 |
-
new_start = round((part['end'] or 0) - medium_dur(), 3)
|
115 |
-
part['start'] = max(new_start, 0.0)
|
116 |
-
else:
|
117 |
-
part['start'] = parts[i - 2]['end']
|
118 |
-
if part['end'] is None:
|
119 |
-
no_next_start = i == len(parts) or parts[i]['start'] is None
|
120 |
-
part['end'] = _curr_max_end(part['start'], i) if no_next_start else parts[i]['start']
|
121 |
-
|
122 |
-
words = [dict(start=word['timestamp'][0], end=word['timestamp'][1], word=word['text']) for word in result]
|
123 |
-
replace_none_ts(words)
|
124 |
-
return WhisperResult([words], force_order=True, check_sorted=True)
|
125 |
-
|
126 |
-
|
127 |
-
def fix_timestamp(pipeline_output: List[Dict[str, Any]], audio: np.ndarray, sample_rate: int) -> List[Dict[str, Any]]:
|
128 |
-
result = _fix_timestamp(sample_rate=sample_rate, audio=audio, result=pipeline_output)
|
129 |
-
result.adjust_by_silence(
|
130 |
-
audio,
|
131 |
-
q_levels=20,
|
132 |
-
k_size=5,
|
133 |
-
sample_rate=sample_rate,
|
134 |
-
min_word_dur=None,
|
135 |
-
word_level=True,
|
136 |
-
verbose=True,
|
137 |
-
nonspeech_error=0.1,
|
138 |
-
use_word_position=True
|
139 |
-
)
|
140 |
-
if result.has_words:
|
141 |
-
result.regroup(True)
|
142 |
-
return [{"timestamp": [s.start, s.end], "text": s.text} for s in result.segments]
|
143 |
-
|
144 |
-
|
145 |
def format_time(start: Optional[float], end: Optional[float]):
|
146 |
|
147 |
def _format_time(seconds: Optional[float]):
|
@@ -157,17 +54,11 @@ def format_time(start: Optional[float], end: Optional[float]):
|
|
157 |
return f"[{_format_time(start)}-> {_format_time(end)}]:"
|
158 |
|
159 |
|
160 |
-
def get_prediction(inputs, prompt: Optional[str]
|
161 |
generate_kwargs = {"language": "japanese", "task": "transcribe"}
|
162 |
if prompt:
|
163 |
generate_kwargs['prompt_ids'] = pipe.tokenizer.get_prompt_ids(prompt, return_tensors='pt').to(device)
|
164 |
-
array = inputs["array"]
|
165 |
-
sr = inputs["sampling_rate"]
|
166 |
prediction = pipe(inputs, return_timestamps=True, generate_kwargs=generate_kwargs)
|
167 |
-
if stabilize_timestamp:
|
168 |
-
prediction['chunks'] = fix_timestamp(pipeline_output=prediction['chunks'], audio=array, sample_rate=sr)
|
169 |
-
if punctuate_text:
|
170 |
-
prediction['chunks'] = PUNCTUATOR.punctuate(prediction['chunks'])
|
171 |
text = "".join([c['text'] for c in prediction['chunks']])
|
172 |
text_timestamped = "\n".join([
|
173 |
f"{format_time(*c['timestamp'])} {c['text']}" for c in prediction['chunks']
|
@@ -175,14 +66,14 @@ def get_prediction(inputs, prompt: Optional[str], punctuate_text: bool = True, s
|
|
175 |
return text, text_timestamped
|
176 |
|
177 |
|
178 |
-
def transcribe(inputs: str, prompt
|
179 |
if inputs is None:
|
180 |
raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
|
181 |
with open(inputs, "rb") as f:
|
182 |
inputs = f.read()
|
183 |
inputs = ffmpeg_read(inputs, pipe.feature_extractor.sampling_rate)
|
184 |
inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
|
185 |
-
return get_prediction(inputs, prompt
|
186 |
|
187 |
|
188 |
def _return_yt_html_embed(yt_url):
|
@@ -216,7 +107,7 @@ def download_yt_audio(yt_url, filename):
|
|
216 |
raise gr.Error(str(err))
|
217 |
|
218 |
|
219 |
-
def yt_transcribe(yt_url, prompt
|
220 |
html_embed_str = _return_yt_html_embed(yt_url)
|
221 |
with tempfile.TemporaryDirectory() as tmpdirname:
|
222 |
filepath = os.path.join(tmpdirname, "video.mp4")
|
@@ -225,7 +116,7 @@ def yt_transcribe(yt_url, prompt, punctuate_text: bool = True, stabilize_timesta
|
|
225 |
inputs = f.read()
|
226 |
inputs = ffmpeg_read(inputs, pipe.feature_extractor.sampling_rate)
|
227 |
inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
|
228 |
-
text, text_timestamped = get_prediction(inputs, prompt
|
229 |
return html_embed_str, text, text_timestamped
|
230 |
|
231 |
|
@@ -235,8 +126,6 @@ mf_transcribe = gr.Interface(
|
|
235 |
inputs=[
|
236 |
gr.inputs.Audio(source="microphone", type="filepath", optional=True),
|
237 |
gr.inputs.Textbox(lines=1, placeholder="Prompt", optional=True),
|
238 |
-
gr.inputs.Checkbox(default=True, label="Add punctuations"),
|
239 |
-
gr.inputs.Checkbox(default=True, label="Stabilize timestamp")
|
240 |
],
|
241 |
outputs=["text", "text"],
|
242 |
layout="horizontal",
|
@@ -251,8 +140,6 @@ file_transcribe = gr.Interface(
|
|
251 |
inputs=[
|
252 |
gr.inputs.Audio(source="upload", type="filepath", optional=True, label="Audio file"),
|
253 |
gr.inputs.Textbox(lines=1, placeholder="Prompt", optional=True),
|
254 |
-
gr.inputs.Checkbox(default=True, label="Add punctuations"),
|
255 |
-
gr.inputs.Checkbox(default=True, label="Stabilize timestamp")
|
256 |
],
|
257 |
outputs=["text", "text"],
|
258 |
layout="horizontal",
|
@@ -266,8 +153,6 @@ yt_transcribe = gr.Interface(
|
|
266 |
inputs=[
|
267 |
gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
|
268 |
gr.inputs.Textbox(lines=1, placeholder="Prompt", optional=True),
|
269 |
-
gr.inputs.Checkbox(default=True, label="Add punctuations"),
|
270 |
-
gr.inputs.Checkbox(default=True, label="Stabilize timestamp")
|
271 |
],
|
272 |
outputs=["html", "text", "text"],
|
273 |
layout="horizontal",
|
|
|
7 |
import torch
|
8 |
import gradio as gr
|
9 |
import yt_dlp as youtube_dl
|
|
|
10 |
from transformers import pipeline
|
11 |
from transformers.pipelines.audio_utils import ffmpeg_read
|
|
|
|
|
12 |
|
13 |
|
14 |
# configuration
|
15 |
+
MODEL_NAME = "kotoba-tech/kotoba-whisper-v1.1"
|
16 |
BATCH_SIZE = 16
|
17 |
CHUNK_LENGTH_S = 15
|
18 |
FILE_LIMIT_MB = 1000
|
19 |
YT_LENGTH_LIMIT_S = 3600 # limit to 1 hour YouTube files
|
|
|
|
|
20 |
# device setting
|
21 |
if torch.cuda.is_available():
|
22 |
torch_dtype = torch.bfloat16
|
|
|
26 |
torch_dtype = torch.float32
|
27 |
device = "cpu"
|
28 |
model_kwargs = {}
|
|
|
29 |
# define the pipeline
|
30 |
pipe = pipeline(
|
|
|
31 |
model=MODEL_NAME,
|
32 |
chunk_length_s=CHUNK_LENGTH_S,
|
33 |
batch_size=BATCH_SIZE,
|
34 |
torch_dtype=torch_dtype,
|
35 |
device=device,
|
36 |
+
model_kwargs=model_kwargs,
|
37 |
+
punctuator=True,
|
38 |
+
stable_ts=True,
|
39 |
)
|
40 |
|
41 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
def format_time(start: Optional[float], end: Optional[float]):
|
43 |
|
44 |
def _format_time(seconds: Optional[float]):
|
|
|
54 |
return f"[{_format_time(start)}-> {_format_time(end)}]:"
|
55 |
|
56 |
|
57 |
+
def get_prediction(inputs, prompt: Optional[str]):
|
58 |
generate_kwargs = {"language": "japanese", "task": "transcribe"}
|
59 |
if prompt:
|
60 |
generate_kwargs['prompt_ids'] = pipe.tokenizer.get_prompt_ids(prompt, return_tensors='pt').to(device)
|
|
|
|
|
61 |
prediction = pipe(inputs, return_timestamps=True, generate_kwargs=generate_kwargs)
|
|
|
|
|
|
|
|
|
62 |
text = "".join([c['text'] for c in prediction['chunks']])
|
63 |
text_timestamped = "\n".join([
|
64 |
f"{format_time(*c['timestamp'])} {c['text']}" for c in prediction['chunks']
|
|
|
66 |
return text, text_timestamped
|
67 |
|
68 |
|
69 |
+
def transcribe(inputs: str, prompt):
|
70 |
if inputs is None:
|
71 |
raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
|
72 |
with open(inputs, "rb") as f:
|
73 |
inputs = f.read()
|
74 |
inputs = ffmpeg_read(inputs, pipe.feature_extractor.sampling_rate)
|
75 |
inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
|
76 |
+
return get_prediction(inputs, prompt)
|
77 |
|
78 |
|
79 |
def _return_yt_html_embed(yt_url):
|
|
|
107 |
raise gr.Error(str(err))
|
108 |
|
109 |
|
110 |
+
def yt_transcribe(yt_url, prompt):
|
111 |
html_embed_str = _return_yt_html_embed(yt_url)
|
112 |
with tempfile.TemporaryDirectory() as tmpdirname:
|
113 |
filepath = os.path.join(tmpdirname, "video.mp4")
|
|
|
116 |
inputs = f.read()
|
117 |
inputs = ffmpeg_read(inputs, pipe.feature_extractor.sampling_rate)
|
118 |
inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
|
119 |
+
text, text_timestamped = get_prediction(inputs, prompt)
|
120 |
return html_embed_str, text, text_timestamped
|
121 |
|
122 |
|
|
|
126 |
inputs=[
|
127 |
gr.inputs.Audio(source="microphone", type="filepath", optional=True),
|
128 |
gr.inputs.Textbox(lines=1, placeholder="Prompt", optional=True),
|
|
|
|
|
129 |
],
|
130 |
outputs=["text", "text"],
|
131 |
layout="horizontal",
|
|
|
140 |
inputs=[
|
141 |
gr.inputs.Audio(source="upload", type="filepath", optional=True, label="Audio file"),
|
142 |
gr.inputs.Textbox(lines=1, placeholder="Prompt", optional=True),
|
|
|
|
|
143 |
],
|
144 |
outputs=["text", "text"],
|
145 |
layout="horizontal",
|
|
|
153 |
inputs=[
|
154 |
gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
|
155 |
gr.inputs.Textbox(lines=1, placeholder="Prompt", optional=True),
|
|
|
|
|
156 |
],
|
157 |
outputs=["html", "text", "text"],
|
158 |
layout="horizontal",
|