Audio: MFCC: Add Mel log spectra output mode when num_ceps is zero

singalsu · lgirdwood · commit 3cd6a6d3c9bf · 2026-05-11T16:27:52.000+01:00
Add a mode where cepstral coefficients are not computed and the
Mel frequency logarithm values are passed directly to the sink
buffer. The mode is activated when sof_mfcc_config member num_ceps
is set to zero.

When num_ceps is zero:
- DCT matrix and cepstral lifter are not allocated or initialized
- The Mel log spectra (num_mel_bins values) are output to the sink
  instead of cepstral coefficients
- A mel_only flag is added to mfcc_state for runtime path selection

This is useful for applications that need Mel spectrogram features
without the DCT transform, such as some neural network audio
front-ends.

Signed-off-by: Seppo Ingalsuo &lt;seppo.ingalsuo@linux.intel.com&gt;
diff --git a/src/audio/mfcc/mfcc_common.c b/src/audio/mfcc/mfcc_common.c
@@ -119,16 +119,22 @@ static int mfcc_stft_process(const struct comp_dev *dev, struct mfcc_state *stat
 					    state->mel_spectra->data, mel_scale_shift);
 #endif
 
-		/* Multiply Mel spectra with DCT matrix to get cepstral coefficients */
-		mat_init_16b(state->cepstral_coef, 1, state->dct.num_out, 7); /* Q8.7 */
-		mat_multiply(state->mel_spectra, state->dct.matrix, state->cepstral_coef);
-
-		/* Apply cepstral lifter */
-		if (state->lifter.cepstral_lifter != 0)
-			mat_multiply_elementwise(state->cepstral_coef, state->lifter.matrix,
-						 state->cepstral_coef);
-
-		cc_count += state->dct.num_out;
+		if (state->mel_only) {
+			/* In Mel-only mode output Mel log spectra directly */
+			cc_count += state->dct.num_in;
+		} else {
+			/* Multiply Mel spectra with DCT matrix to get cepstral coefficients */
+			mat_init_16b(state->cepstral_coef, 1, state->dct.num_out, 7); /* Q8.7 */
+			mat_multiply(state->mel_spectra, state->dct.matrix, state->cepstral_coef);
+
+			/* Apply cepstral lifter */
+			if (state->lifter.cepstral_lifter != 0) {
+				mat_multiply_elementwise(state->cepstral_coef, state->lifter.matrix,
+							 state->cepstral_coef);
+			}
+
+			cc_count += state->dct.num_out;
+		}
 
 		/* Output to sink buffer */
 	}
@@ -205,9 +211,17 @@ void mfcc_s16_default(struct processing_module *mod, struct input_stream_buffer
 	 */
 	zero_samples = frames * audio_stream_get_channels(sink);
 	if (num_ceps > 0) {
+		int16_t *out_data;
+
+		if (state->mel_only) {
+			out_data = state->mel_spectra->data;
+		} else {
+			out_data = state->cepstral_coef->data;
+		}
+
 		zero_samples -= num_ceps + num_magic;
 		w_ptr = mfcc_sink_copy_data_s16(sink, w_ptr, num_magic, (int16_t *)&magic);
-		w_ptr = mfcc_sink_copy_data_s16(sink, w_ptr, num_ceps, state->cepstral_coef->data);
+		w_ptr = mfcc_sink_copy_data_s16(sink, w_ptr, num_ceps, out_data);
 	}
 
 	w_ptr = mfcc_sink_copy_zero_s16(sink, w_ptr, zero_samples);
diff --git a/src/audio/mfcc/mfcc_setup.c b/src/audio/mfcc/mfcc_setup.c
@@ -248,23 +248,37 @@ int mfcc_setup(struct processing_module *mod, int max_frames, int sample_rate, i
 		goto free_fft_out;
 	}
 
-	/* Setup DCT */
-	dct->num_in = config->num_mel_bins;
-	dct->num_out = config->num_ceps;
-	dct->type = (enum dct_type)config->dct;
-	dct->ortho = true;
-	ret = mod_dct_initialize_16(mod, dct);
-	if (ret < 0) {
-		comp_err(dev, "Failed DCT init");
-		goto free_melfb_data;
-	}
-
-	state->lifter.num_ceps = config->num_ceps;
-	state->lifter.cepstral_lifter = config->cepstral_lifter; /* Q7.9 max 64.0*/
-	ret = mfcc_get_cepstral_lifter(mod, &state->lifter);
-	if (ret < 0) {
-		comp_err(dev, "Failed cepstral lifter");
-		goto free_dct_matrix;
+	/* Setup DCT and cepstral lifter only when num_ceps > 0.
+	 * When num_ceps is zero, skip DCT/lifter and output Mel
+	 * log spectra directly.
+	 */
+	if (config->num_ceps > 0) {
+		dct->num_in = config->num_mel_bins;
+		dct->num_out = config->num_ceps;
+		dct->type = (enum dct_type)config->dct;
+		dct->ortho = true;
+		ret = mod_dct_initialize_16(mod, dct);
+		if (ret < 0) {
+			comp_err(dev, "Failed DCT init");
+			goto free_melfb_data;
+		}
+
+		state->lifter.num_ceps = config->num_ceps;
+		state->lifter.cepstral_lifter = config->cepstral_lifter; /* Q7.9 max 64.0*/
+		ret = mfcc_get_cepstral_lifter(mod, &state->lifter);
+		if (ret < 0) {
+			comp_err(dev, "Failed cepstral lifter");
+			goto free_dct_matrix;
+		}
+
+		state->mel_only = false;
+	} else {
+		comp_info(dev, "num_ceps is 0, Mel log spectra output mode");
+		dct->num_in = config->num_mel_bins;
+		dct->num_out = 0;
+		dct->matrix = NULL;
+		state->lifter.matrix = NULL;
+		state->mel_only = true;
 	}
 
 	/* Scratch overlay during runtime
@@ -288,8 +302,12 @@ int mfcc_setup(struct processing_module *mod, int max_frames, int sample_rate, i
 	/* Use FFT buffer as scratch for later computed data */
 	state->power_spectra = (int32_t *)&fft->fft_buf[0];
 	state->mel_spectra = (struct mat_matrix_16b *)&fft->fft_out[0];
-	state->cepstral_coef = (struct mat_matrix_16b *)
-		&state->mel_spectra->data[state->dct.num_in];
+	if (!state->mel_only) {
+		state->cepstral_coef =
+			(struct mat_matrix_16b *)&state->mel_spectra->data[state->dct.num_in];
+	} else {
+		state->cepstral_coef = NULL;
+	}
 
 	/* Set initial state for STFT */
 	state->waiting_fill = true;
diff --git a/src/include/sof/audio/mfcc/mfcc_comp.h b/src/include/sof/audio/mfcc/mfcc_comp.h
@@ -125,6 +125,7 @@ struct mfcc_state {
 	int low_freq;
 	int high_freq;
 	int sample_rate;
+	bool mel_only; /**< When true, output Mel spectra instead of cepstral coefficients */
 	bool waiting_fill; /**< booleans */
 	bool prev_samples_valid;
 	size_t sample_buffers_size; /**< bytes */
