SillyTavern/SillyTavern-extras
1
0
Fork 0
mirror of https://github.com/SillyTavern/SillyTavern-Extras.git synced 2026-05-01 11:51:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

change default computing types for different devices

Browse Source
This commit is contained in:
ArrangingFear56
2024-12-10 01:23:13 +08:00
parent d86a8622c3
commit 6435ea40f1
2 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
modules/speech_recognition/faster_whisper_module.py
View File

@@ -22,17 +22,22 @@ RECORDING_FILE_PATH = "stt_test.wav"
model_size = "large-v3-turbo" model_size = "large-v3-turbo"
def load_model(file_path=None,whisper_device="cuda",whisper_compute_type='float16'): def load_model(file_path=None,whisper_device="cuda",whisper_compute_type="auto"):
""" """
Load given vosk model from file or default to en-us model. Load given vosk model from file or default to en-us model.
Download model to user cache folder, example: C:/Users/toto/.cache/vosk Download model to user cache folder, example: C:/Users/toto/.cache/vosk
""" """
if whisper_compute_type=="auto":
whisper_compute_type=(
"int8"
if whisper_device=="cpu"
else "float16")
if file_path is None: if file_path is None:
print(f"faster-whisper using {model_size}") print(f"faster-whisper using {model_size} model with {whisper_compute_type}")
return WhisperModel(model_size, device=whisper_device, compute_type=whisper_compute_type) return WhisperModel(model_size, device=whisper_device, compute_type=whisper_compute_type)
else: else:
print(f"faster-whisper using {file_path}") print(f"faster-whisper using {file_path} model with {whisper_compute_type}")
return WhisperModel(file_path, device=whisper_device, compute_type=whisper_compute_type) return WhisperModel(file_path, device=whisper_device, compute_type=whisper_compute_type)
def process_audio(): def process_audio():

4
server.py
View File

@@ -939,7 +939,7 @@ parser.add_argument("--stt-whisper-model-path", help="Load a custom whisper spee
parser.add_argument("--use-faster-whisper", action="store_true", help="Choose to use faster-whisper instead of whisper") parser.add_argument("--use-faster-whisper", action="store_true", help="Choose to use faster-whisper instead of whisper")
parser.add_argument("--faster-whisper-cpu", action="store_true", help="Use cpu to run faster-whisper, saves VRAM but much slower") parser.add_argument("--faster-whisper-cpu", action="store_true", help="Use cpu to run faster-whisper, saves VRAM but much slower")
parser.add_argument("--faster-whisper-type", help="Choose faster-whisper compute type, defaults to float16") parser.add_argument("--faster-whisper-type", help="Choose faster-whisper compute type, defaults to float16 for cuda and int8 for cpu")
# sd_group = parser.add_mutually_exclusive_group() # sd_group = parser.add_mutually_exclusive_group()
@@ -1186,7 +1186,7 @@ if "whisper-stt" in modules:
faster_whisper_type=( faster_whisper_type=(
args.faster_whisper_type args.faster_whisper_type
if args.faster_whisper_type if args.faster_whisper_type
else "float16") else "auto")
print(f"Initializing Faster-Whisper speech-recognition (from ST request file) on {faster_whisper_device}") print(f"Initializing Faster-Whisper speech-recognition (from ST request file) on {faster_whisper_device}")
import modules.speech_recognition.faster_whisper_module as whisper_module import modules.speech_recognition.faster_whisper_module as whisper_module