Spaces:
Running
Running
jhj0517
commited on
Commit
·
21bbf6d
1
Parent(s):
0da25b6
Update visibility by whisper implementation
Browse files- modules/whisper/data_classes.py +111 -101
modules/whisper/data_classes.py
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
from typing import Optional, Dict, List
|
4 |
-
from pydantic import BaseModel, Field, field_validator
|
5 |
from gradio_i18n import Translate, gettext as _
|
6 |
from enum import Enum
|
7 |
from copy import deepcopy
|
@@ -17,6 +17,8 @@ class WhisperImpl(Enum):
|
|
17 |
|
18 |
|
19 |
class BaseParams(BaseModel):
|
|
|
|
|
20 |
def to_dict(self) -> Dict:
|
21 |
return self.model_dump()
|
22 |
|
@@ -231,7 +233,6 @@ class WhisperParams(BaseParams):
|
|
231 |
gt=0,
|
232 |
description="Threshold for gzip compression ratio"
|
233 |
)
|
234 |
-
batch_size: int = Field(default=24, gt=0, description="Batch size for processing")
|
235 |
length_penalty: float = Field(default=1.0, gt=0, description="Exponential length penalty")
|
236 |
repetition_penalty: float = Field(default=1.0, gt=0, description="Penalty for repeated tokens")
|
237 |
no_repeat_ngram_size: int = Field(default=0, ge=0, description="Size of n-grams to prevent repetition")
|
@@ -271,6 +272,7 @@ class WhisperParams(BaseParams):
|
|
271 |
gt=0,
|
272 |
description="Number of segments for language detection"
|
273 |
)
|
|
|
274 |
|
275 |
@field_validator('lang')
|
276 |
def validate_lang(cls, v):
|
@@ -375,108 +377,116 @@ class WhisperParams(BaseParams):
|
|
375 |
info="Threshold for gzip compression ratio"
|
376 |
)
|
377 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
378 |
if whisper_type == WhisperImpl.FASTER_WHISPER:
|
379 |
-
|
380 |
-
|
381 |
-
label="Length Penalty",
|
382 |
-
value=defaults.get("length_penalty", cls.__fields__["length_penalty"].default),
|
383 |
-
info="Exponential length penalty",
|
384 |
-
visible=whisper_type == "faster_whisper"
|
385 |
-
),
|
386 |
-
gr.Number(
|
387 |
-
label="Repetition Penalty",
|
388 |
-
value=defaults.get("repetition_penalty", cls.__fields__["repetition_penalty"].default),
|
389 |
-
info="Penalty for repeated tokens"
|
390 |
-
),
|
391 |
-
gr.Number(
|
392 |
-
label="No Repeat N-gram Size",
|
393 |
-
value=defaults.get("no_repeat_ngram_size", cls.__fields__["no_repeat_ngram_size"].default),
|
394 |
-
precision=0,
|
395 |
-
info="Size of n-grams to prevent repetition"
|
396 |
-
),
|
397 |
-
gr.Textbox(
|
398 |
-
label="Prefix",
|
399 |
-
value=defaults.get("prefix", cls.__fields__["prefix"].default),
|
400 |
-
info="Prefix text for first window"
|
401 |
-
),
|
402 |
-
gr.Checkbox(
|
403 |
-
label="Suppress Blank",
|
404 |
-
value=defaults.get("suppress_blank", cls.__fields__["suppress_blank"].default),
|
405 |
-
info="Suppress blank outputs at start of sampling"
|
406 |
-
),
|
407 |
-
gr.Textbox(
|
408 |
-
label="Suppress Tokens",
|
409 |
-
value=defaults.get("suppress_tokens", cls.__fields__["suppress_tokens"].default),
|
410 |
-
info="Token IDs to suppress"
|
411 |
-
),
|
412 |
-
gr.Number(
|
413 |
-
label="Max Initial Timestamp",
|
414 |
-
value=defaults.get("max_initial_timestamp", cls.__fields__["max_initial_timestamp"].default),
|
415 |
-
info="Maximum initial timestamp"
|
416 |
-
),
|
417 |
-
gr.Checkbox(
|
418 |
-
label="Word Timestamps",
|
419 |
-
value=defaults.get("word_timestamps", cls.__fields__["word_timestamps"].default),
|
420 |
-
info="Extract word-level timestamps"
|
421 |
-
),
|
422 |
-
gr.Textbox(
|
423 |
-
label="Prepend Punctuations",
|
424 |
-
value=defaults.get("prepend_punctuations", cls.__fields__["prepend_punctuations"].default),
|
425 |
-
info="Punctuations to merge with next word"
|
426 |
-
),
|
427 |
-
gr.Textbox(
|
428 |
-
label="Append Punctuations",
|
429 |
-
value=defaults.get("append_punctuations", cls.__fields__["append_punctuations"].default),
|
430 |
-
info="Punctuations to merge with previous word"
|
431 |
-
),
|
432 |
-
gr.Number(
|
433 |
-
label="Max New Tokens",
|
434 |
-
value=defaults.get("max_new_tokens", cls.__fields__["max_new_tokens"].default),
|
435 |
-
precision=0,
|
436 |
-
info="Maximum number of new tokens per chunk"
|
437 |
-
),
|
438 |
-
gr.Number(
|
439 |
-
label="Chunk Length (s)",
|
440 |
-
value=defaults.get("chunk_length", cls.__fields__["chunk_length"].default),
|
441 |
-
precision=0,
|
442 |
-
info="Length of audio segments in seconds"
|
443 |
-
),
|
444 |
-
gr.Number(
|
445 |
-
label="Hallucination Silence Threshold (sec)",
|
446 |
-
value=defaults.get("hallucination_silence_threshold",
|
447 |
-
cls.__fields__["hallucination_silence_threshold"].default),
|
448 |
-
info="Threshold for skipping silent periods in hallucination detection"
|
449 |
-
),
|
450 |
-
gr.Textbox(
|
451 |
-
label="Hotwords",
|
452 |
-
value=defaults.get("hotwords", cls.__fields__["hotwords"].default),
|
453 |
-
info="Hotwords/hint phrases for the model"
|
454 |
-
),
|
455 |
-
gr.Number(
|
456 |
-
label="Language Detection Threshold",
|
457 |
-
value=defaults.get("language_detection_threshold",
|
458 |
-
cls.__fields__["language_detection_threshold"].default),
|
459 |
-
info="Threshold for language detection probability"
|
460 |
-
),
|
461 |
-
gr.Number(
|
462 |
-
label="Language Detection Segments",
|
463 |
-
value=defaults.get("language_detection_segments",
|
464 |
-
cls.__fields__["language_detection_segments"].default),
|
465 |
-
precision=0,
|
466 |
-
info="Number of segments for language detection"
|
467 |
-
)
|
468 |
-
]
|
469 |
|
470 |
if whisper_type == WhisperImpl.INSANELY_FAST_WHISPER:
|
471 |
-
|
472 |
-
|
473 |
-
|
474 |
-
|
475 |
-
|
476 |
-
info="Batch size for processing",
|
477 |
-
visible=whisper_type == "insanely_fast_whisper"
|
478 |
-
)
|
479 |
-
]
|
480 |
return inputs
|
481 |
|
482 |
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
from typing import Optional, Dict, List
|
4 |
+
from pydantic import BaseModel, Field, field_validator, ConfigDict
|
5 |
from gradio_i18n import Translate, gettext as _
|
6 |
from enum import Enum
|
7 |
from copy import deepcopy
|
|
|
17 |
|
18 |
|
19 |
class BaseParams(BaseModel):
|
20 |
+
model_config = ConfigDict(protected_namespaces=())
|
21 |
+
|
22 |
def to_dict(self) -> Dict:
|
23 |
return self.model_dump()
|
24 |
|
|
|
233 |
gt=0,
|
234 |
description="Threshold for gzip compression ratio"
|
235 |
)
|
|
|
236 |
length_penalty: float = Field(default=1.0, gt=0, description="Exponential length penalty")
|
237 |
repetition_penalty: float = Field(default=1.0, gt=0, description="Penalty for repeated tokens")
|
238 |
no_repeat_ngram_size: int = Field(default=0, ge=0, description="Size of n-grams to prevent repetition")
|
|
|
272 |
gt=0,
|
273 |
description="Number of segments for language detection"
|
274 |
)
|
275 |
+
batch_size: int = Field(default=24, gt=0, description="Batch size for processing")
|
276 |
|
277 |
@field_validator('lang')
|
278 |
def validate_lang(cls, v):
|
|
|
377 |
info="Threshold for gzip compression ratio"
|
378 |
)
|
379 |
]
|
380 |
+
|
381 |
+
faster_whisper_inputs = [
|
382 |
+
gr.Number(
|
383 |
+
label="Length Penalty",
|
384 |
+
value=defaults.get("length_penalty", cls.__fields__["length_penalty"].default),
|
385 |
+
info="Exponential length penalty",
|
386 |
+
),
|
387 |
+
gr.Number(
|
388 |
+
label="Repetition Penalty",
|
389 |
+
value=defaults.get("repetition_penalty", cls.__fields__["repetition_penalty"].default),
|
390 |
+
info="Penalty for repeated tokens"
|
391 |
+
),
|
392 |
+
gr.Number(
|
393 |
+
label="No Repeat N-gram Size",
|
394 |
+
value=defaults.get("no_repeat_ngram_size", cls.__fields__["no_repeat_ngram_size"].default),
|
395 |
+
precision=0,
|
396 |
+
info="Size of n-grams to prevent repetition"
|
397 |
+
),
|
398 |
+
gr.Textbox(
|
399 |
+
label="Prefix",
|
400 |
+
value=defaults.get("prefix", cls.__fields__["prefix"].default),
|
401 |
+
info="Prefix text for first window"
|
402 |
+
),
|
403 |
+
gr.Checkbox(
|
404 |
+
label="Suppress Blank",
|
405 |
+
value=defaults.get("suppress_blank", cls.__fields__["suppress_blank"].default),
|
406 |
+
info="Suppress blank outputs at start of sampling"
|
407 |
+
),
|
408 |
+
gr.Textbox(
|
409 |
+
label="Suppress Tokens",
|
410 |
+
value=defaults.get("suppress_tokens", cls.__fields__["suppress_tokens"].default),
|
411 |
+
info="Token IDs to suppress"
|
412 |
+
),
|
413 |
+
gr.Number(
|
414 |
+
label="Max Initial Timestamp",
|
415 |
+
value=defaults.get("max_initial_timestamp", cls.__fields__["max_initial_timestamp"].default),
|
416 |
+
info="Maximum initial timestamp"
|
417 |
+
),
|
418 |
+
gr.Checkbox(
|
419 |
+
label="Word Timestamps",
|
420 |
+
value=defaults.get("word_timestamps", cls.__fields__["word_timestamps"].default),
|
421 |
+
info="Extract word-level timestamps"
|
422 |
+
),
|
423 |
+
gr.Textbox(
|
424 |
+
label="Prepend Punctuations",
|
425 |
+
value=defaults.get("prepend_punctuations", cls.__fields__["prepend_punctuations"].default),
|
426 |
+
info="Punctuations to merge with next word"
|
427 |
+
),
|
428 |
+
gr.Textbox(
|
429 |
+
label="Append Punctuations",
|
430 |
+
value=defaults.get("append_punctuations", cls.__fields__["append_punctuations"].default),
|
431 |
+
info="Punctuations to merge with previous word"
|
432 |
+
),
|
433 |
+
gr.Number(
|
434 |
+
label="Max New Tokens",
|
435 |
+
value=defaults.get("max_new_tokens", cls.__fields__["max_new_tokens"].default),
|
436 |
+
precision=0,
|
437 |
+
info="Maximum number of new tokens per chunk"
|
438 |
+
),
|
439 |
+
gr.Number(
|
440 |
+
label="Chunk Length (s)",
|
441 |
+
value=defaults.get("chunk_length", cls.__fields__["chunk_length"].default),
|
442 |
+
precision=0,
|
443 |
+
info="Length of audio segments in seconds"
|
444 |
+
),
|
445 |
+
gr.Number(
|
446 |
+
label="Hallucination Silence Threshold (sec)",
|
447 |
+
value=defaults.get("hallucination_silence_threshold",
|
448 |
+
cls.__fields__["hallucination_silence_threshold"].default),
|
449 |
+
info="Threshold for skipping silent periods in hallucination detection"
|
450 |
+
),
|
451 |
+
gr.Textbox(
|
452 |
+
label="Hotwords",
|
453 |
+
value=defaults.get("hotwords", cls.__fields__["hotwords"].default),
|
454 |
+
info="Hotwords/hint phrases for the model"
|
455 |
+
),
|
456 |
+
gr.Number(
|
457 |
+
label="Language Detection Threshold",
|
458 |
+
value=defaults.get("language_detection_threshold",
|
459 |
+
cls.__fields__["language_detection_threshold"].default),
|
460 |
+
info="Threshold for language detection probability"
|
461 |
+
),
|
462 |
+
gr.Number(
|
463 |
+
label="Language Detection Segments",
|
464 |
+
value=defaults.get("language_detection_segments",
|
465 |
+
cls.__fields__["language_detection_segments"].default),
|
466 |
+
precision=0,
|
467 |
+
info="Number of segments for language detection"
|
468 |
+
)
|
469 |
+
]
|
470 |
+
|
471 |
+
insanely_fast_whisper_inputs = [
|
472 |
+
gr.Number(
|
473 |
+
label="Batch Size",
|
474 |
+
value=defaults.get("batch_size", cls.__fields__["batch_size"].default),
|
475 |
+
precision=0,
|
476 |
+
info="Batch size for processing"
|
477 |
+
)
|
478 |
+
]
|
479 |
+
|
480 |
if whisper_type == WhisperImpl.FASTER_WHISPER:
|
481 |
+
for input_component in faster_whisper_inputs:
|
482 |
+
input_component.visible = True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
483 |
|
484 |
if whisper_type == WhisperImpl.INSANELY_FAST_WHISPER:
|
485 |
+
for input_component in insanely_fast_whisper_inputs:
|
486 |
+
input_component.visible = True
|
487 |
+
|
488 |
+
inputs += faster_whisper_inputs + insanely_fast_whisper_inputs
|
489 |
+
|
|
|
|
|
|
|
|
|
490 |
return inputs
|
491 |
|
492 |
|