speech_features.py

# speech_features.py
import torch
import opensmile
import librosa

# Feature extraction and silence detection
def load_silero_vad():
    vad_model, vad_utils = torch.hub.load(
        repo_or_dir="snakers4/silero-vad",
        model="silero_vad",
        trust_repo=True
    )
    return vad_model, vad_utils

def get_speech_timestamps(wav, vad_model, sampling_rate=16000):
    (get_speech_timestamps, _, _, _, _) = vad_utils
    return get_speech_timestamps(wav, vad_model, sampling_rate)

def compute_pitch_energy(y, sr, HOP=512, FRAME_LENGTH=2048):
    # pitch & energy calculation (using librosa)
    f0, _, _ = librosa.pyin(y, fmin=70, fmax=400, sr=sr, hop_length=HOP)
    pitch_mean = float(np.nanmean(f0)) if np.any(~np.isnan(f0)) else float("nan")
    pitch_std  = float(np.nanstd(f0))  if np.any(~np.isnan(f0)) else float("nan")

    rms = librosa.feature.rms(y=y, frame_length=FRAME_LENGTH, hop_length=HOP)[0]
    energy_mean = float(np.mean(rms))
    energy_std  = float(np.std(rms))
    
    return pitch_mean, pitch_std, energy_mean, energy_std, f0, rms