70 lines
2.1 KiB
Python
70 lines
2.1 KiB
Python
|
import logging
|
||
|
|
|
||
|
|
from faster_whisper import WhisperModel
|
||
|
|
|
||
|
|
logging.basicConfig(level=logging.INFO)
|
||
|
|
logger = logging.getLogger(__name__)
|
||
|
|
|
||
|
|
|
||
|
|
class Transcriber:
|
||
|
|
"""
|
||
|
|
Converts audio chunks (numpy arrays) into text using faster-whisper.
|
||
|
|
"""
|
||
|
|
|
||
|
|
def __init__(self, model_size="base", device="cpu", compute_type="int8"):
|
||
|
|
"""
|
||
|
|
Initializes the faster-whisper model.
|
||
|
|
|
||
|
|
Args:
|
||
|
|
model_size (str): The size of the model to use (e.g., "tiny", "base", "small").
|
||
|
|
device (str): The device to run the model on ("cpu" or "cuda").
|
||
|
|
compute_type (str): The compute type to use (e.g., "int8", "float16").
|
||
|
|
"""
|
||
|
|
logger.info(
|
||
|
|
f"Loading faster-whisper model: {model_size} on {device} ({compute_type})..."
|
||
|
|
)
|
||
|
|
try:
|
||
|
|
self.model = WhisperModel(
|
||
|
|
model_size, device=device, compute_type=compute_type
|
||
|
|
)
|
||
|
|
logger.info("Model loaded successfully.")
|
||
|
|
except Exception as e:
|
||
|
|
logger.error(f"Failed to load faster-whisper model: {e}")
|
||
|
|
raise
|
||
|
|
|
||
|
|
def transcribe(self, audio_chunk):
|
||
|
|
"""
|
||
|
|
Transcribes a single audio chunk.
|
||
|
|
|
||
|
|
Args:
|
||
|
|
audio_chunk (np.ndarray): The audio data as a numpy array.
|
||
|
|
|
||
|
|
Returns:
|
||
|
|
str: The transcribed text.
|
||
|
|
"""
|
||
|
|
if audio_chunk is None:
|
||
|
|
return ""
|
||
|
|
|
||
|
|
try:
|
||
|
|
# faster-whisper expects audio in float32
|
||
|
|
audio_data = audio_chunk.astype("float32")
|
||
|
|
|
||
|
|
# Transcribe the audio
|
||
|
|
segments, info = self.model.transcribe(audio_data, beam_size=5)
|
||
|
|
|
||
|
|
# Combine segments into a single string
|
||
|
|
text = " ".join([segment.text.strip() for segment in segments])
|
||
|
|
|
||
|
|
return text.strip()
|
||
|
|
except Exception as e:
|
||
|
|
logger.error(f"Transcription error: {e}")
|
||
|
|
return ""
|
||
|
|
|
||
|
|
def close(self):
|
||
|
|
"""
|
||
|
|
Explicitly release model resources if necessary.
|
||
|
|
"""
|
||
|
|
# faster-whisper's WhisperModel doesn't have a standard close(),
|
||
|
|
# but we'll provide this for consistency.
|
||
|
|
pass
|