Spaces:
Running
Running
jhj0517
commited on
Commit
·
87272f5
1
Parent(s):
292ccb4
add `format_result()`
Browse files
modules/insanely_fast_whisper_inference.py
CHANGED
|
@@ -1,20 +1,24 @@
|
|
| 1 |
-
import whisper
|
| 2 |
-
import gradio as gr
|
| 3 |
-
import time
|
| 4 |
import os
|
| 5 |
-
|
| 6 |
import numpy as np
|
|
|
|
| 7 |
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
-
from modules.whisper_base import WhisperBase
|
| 10 |
from modules.whisper_parameter import *
|
|
|
|
| 11 |
|
| 12 |
|
| 13 |
class InsanelyFastWhisperInference(WhisperBase):
|
| 14 |
def __init__(self):
|
| 15 |
super().__init__(
|
| 16 |
-
model_dir=os.path.join("models", "Whisper")
|
| 17 |
)
|
|
|
|
| 18 |
|
| 19 |
def transcribe(self,
|
| 20 |
audio: Union[str, np.ndarray, torch.Tensor],
|
|
@@ -52,21 +56,14 @@ class InsanelyFastWhisperInference(WhisperBase):
|
|
| 52 |
def progress_callback(progress_value):
|
| 53 |
progress(progress_value, desc="Transcribing..")
|
| 54 |
|
| 55 |
-
segments_result = self.model
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
fp16=True if params.compute_type == "float16" else False,
|
| 63 |
-
best_of=params.best_of,
|
| 64 |
-
patience=params.patience,
|
| 65 |
-
temperature=params.temperature,
|
| 66 |
-
compression_ratio_threshold=params.compression_ratio_threshold,
|
| 67 |
-
progress_callback=progress_callback,)["segments"]
|
| 68 |
elapsed_time = time.time() - start_time
|
| 69 |
-
|
| 70 |
return segments_result, elapsed_time
|
| 71 |
|
| 72 |
def update_model(self,
|
|
@@ -90,8 +87,34 @@ class InsanelyFastWhisperInference(WhisperBase):
|
|
| 90 |
progress(0, desc="Initializing Model..")
|
| 91 |
self.current_compute_type = compute_type
|
| 92 |
self.current_model_size = model_size
|
| 93 |
-
|
| 94 |
-
|
|
|
|
|
|
|
|
|
|
| 95 |
device=self.device,
|
| 96 |
-
|
| 97 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
+
import time
|
| 3 |
import numpy as np
|
| 4 |
+
from typing import BinaryIO, Union, Tuple, List
|
| 5 |
import torch
|
| 6 |
+
import transformers
|
| 7 |
+
from transformers import pipeline
|
| 8 |
+
from transformers.utils import is_flash_attn_2_available
|
| 9 |
+
import whisper
|
| 10 |
+
import gradio as gr
|
| 11 |
|
|
|
|
| 12 |
from modules.whisper_parameter import *
|
| 13 |
+
from modules.whisper_base import WhisperBase
|
| 14 |
|
| 15 |
|
| 16 |
class InsanelyFastWhisperInference(WhisperBase):
|
| 17 |
def __init__(self):
|
| 18 |
super().__init__(
|
| 19 |
+
model_dir=os.path.join("models", "Whisper", "insanely_fast_whisper")
|
| 20 |
)
|
| 21 |
+
self.available_compute_types = ["float16"]
|
| 22 |
|
| 23 |
def transcribe(self,
|
| 24 |
audio: Union[str, np.ndarray, torch.Tensor],
|
|
|
|
| 56 |
def progress_callback(progress_value):
|
| 57 |
progress(progress_value, desc="Transcribing..")
|
| 58 |
|
| 59 |
+
segments_result = self.model(
|
| 60 |
+
inputs=audio,
|
| 61 |
+
chunk_length_s=30,
|
| 62 |
+
batch_size=24,
|
| 63 |
+
return_timestamps=True,
|
| 64 |
+
)
|
| 65 |
+
segments_result = self.format_result(transcribed_result=segments_result)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
elapsed_time = time.time() - start_time
|
|
|
|
| 67 |
return segments_result, elapsed_time
|
| 68 |
|
| 69 |
def update_model(self,
|
|
|
|
| 87 |
progress(0, desc="Initializing Model..")
|
| 88 |
self.current_compute_type = compute_type
|
| 89 |
self.current_model_size = model_size
|
| 90 |
+
|
| 91 |
+
self.model = pipeline(
|
| 92 |
+
"automatic-speech-recognition",
|
| 93 |
+
model=os.path.join(self.model_dir, model_size),
|
| 94 |
+
torch_dtype=self.current_compute_type,
|
| 95 |
device=self.device,
|
| 96 |
+
model_kwargs={"attn_implementation": "flash_attention_2"} if is_flash_attn_2_available() else {"attn_implementation": "sdpa"},
|
| 97 |
+
)
|
| 98 |
+
|
| 99 |
+
@staticmethod
|
| 100 |
+
def format_result(transcribed_result: dict) -> List[dict]:
|
| 101 |
+
"""
|
| 102 |
+
Format the transcription result of insanely_fast_whisper as the same with other implementation.
|
| 103 |
+
|
| 104 |
+
Parameters
|
| 105 |
+
----------
|
| 106 |
+
transcribed_result: dict
|
| 107 |
+
Transcription result of the insanely_fast_whisper
|
| 108 |
+
|
| 109 |
+
Returns
|
| 110 |
+
----------
|
| 111 |
+
result: List[dict]
|
| 112 |
+
Formatted result as the same with other implementation
|
| 113 |
+
"""
|
| 114 |
+
result = transcribed_result["chunks"]
|
| 115 |
+
for item in result:
|
| 116 |
+
start, end = item["timestamp"][0], item["timestamp"][1]
|
| 117 |
+
item["start"] = start
|
| 118 |
+
item["end"] = end
|
| 119 |
+
return result
|
| 120 |
+
|