utils/audio_utils.py

import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import librosa
import numpy as np
from pydub import AudioSegment
from config import SPEECH_MODEL, TTS_MODEL

# Initialize models
processor = Wav2Vec2Processor.from_pretrained(SPEECH_MODEL)
model = Wav2Vec2ForCTC.from_pretrained(SPEECH_MODEL)

def speech_to_text(audio_file):
    audio_input, _ = librosa.load(audio_file, sr=16000)
    input_values = processor(audio_input, return_tensors="pt", sampling_rate=16000).input_values
    logits = model(input_values).logits
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)[0]
    return transcription

def text_to_speech(text):
    tts_model = torch.hub.load('snakers4/silero-models', 'silero_tts', model_name=TTS_MODEL)
    audio = tts_model.apply_tts(text=text, speaker='en_0', sample_rate=48000)
    
    # Convert the audio tensor to a numpy array
    audio_np = audio.numpy()
    
    # Normalize the audio to 16-bit PCM range
    audio_np = (audio_np * 32767).astype(np.int16)
    
    # Create an AudioSegment directly from the numpy array
    audio_segment = AudioSegment(
        audio_np.tobytes(),
        frame_rate=48000,
        sample_width=2,
        channels=1
    )
    
    return audio_segment

def process_audio_chunk(chunk):
    # Assuming chunk is a byte string of raw audio data
    audio_np = np.frombuffer(chunk, dtype=np.int16)
    audio_float = audio_np.astype(np.float32) / 32768.0
    return speech_to_text(audio_float)