|TensorFlow 1 version||View source on GitHub|
Computes MFCCs of
Compat aliases for migration
See Migration guide for more details.
tf.signal.mfccs_from_log_mel_spectrograms( log_mel_spectrograms, name=None )
Implemented with GPU-compatible ops and supports gradients.
Mel-Frequency Cepstral Coefficient (MFCC) calculation consists of taking the DCT-II of a log-magnitude mel-scale spectrogram. HTK's MFCCs use a particular scaling of the DCT-II which is almost orthogonal normalization. We follow this convention.
num_mel_bins MFCCs are returned and it is up to the caller to select
a subset of the MFCCs based on their application. For example, it is typical
to only use the first few for speech recognition, as this results in
an approximately pitch-invariant representation of the signal.
batch_size, num_samples, sample_rate = 32, 32000, 16000.0 # A Tensor of [batch_size, num_samples] mono PCM samples in the range [-1, 1]. pcm = tf.random.normal([batch_size, num_samples], dtype=tf.float32) # A 1024-point STFT with frames of 64 ms and 75% overlap. stfts = tf.signal.stft(pcm, frame_length=1024, frame_step=256, fft_length=1024) spectrograms = tf.abs(stfts) # Warp the linear scale spectrograms into the mel-scale. num_spectrogram_bins = stfts.shape[-1].value lower_edge_hertz, upper_edge_hertz, num_mel_bins = 80.0, 7600.0, 80 linear_to_mel_weight_matrix = tf.signal.linear_to_mel_weight_matrix( num_mel_bins, num_spectrogram_bins, sample_rate, lower_edge_hertz, upper_edge_hertz) mel_spectrograms = tf.tensordot( spectrograms, linear_to_mel_weight_matrix, 1) mel_spectrograms.set_shape(spectrograms.shape[:-1].concatenate( linear_to_me