Module features 

Log-mel filterbank (fbank) feature extraction for speaker embeddings.

Typical parameters for ECAPA-TDNN (16 kHz):

• n_fft = 512
• win_length = 400 (25 ms)
• hop_length = 160 (10 ms)
• n_mels = 80
• f_min = 20.0, f_max = 7600.0
• pre_emphasis = 0.97

Structs

FbankConfig

Configuration for log-mel filterbank extraction.

FbankExtractor

Cached log-mel filterbank extractor.

Enums

FbankError

Error during fbank computation.

Functions

apply_cmvn

Apply cepstral mean normalization (CMN) to fbank features.

compute_fbankDeprecated

Standalone log-mel filterbank computation.