Audio: MFCC: More updates and topologies to run MFCC for Mel spectrogram audio features in SDW PCs

scripts/rebuild-testbench.sh

@@ -97,7 +97,7 @@ export_xtensa_setup()
     cat <<EOFSETUP > "$export_script"
 export XTENSA_TOOLS_ROOT=$XTENSA_TOOLS_ROOT
 export XTENSA_CORE=$XTENSA_CORE
-XTENSA_PATH=$tools_bin
+export XTENSA_PATH=$tools_bin
 EOFSETUP
 }
 

src/audio/mfcc/Kconfig

@@ -4,7 +4,7 @@ config COMP_MFCC
 	tristate "MFCC component"
 	depends on COMP_MODULE_ADAPTER
 	select CORDIC_FIXED
-	select MATH_16BIT_MEL_FILTERBANK
+	select MATH_32BIT_MEL_FILTERBANK
 	select MATH_AUDITORY
 	select MATH_DCT
 	select MATH_DECIBELS

src/audio/mfcc/mfcc.c

@@ -160,7 +160,6 @@ static int mfcc_prepare(struct processing_module *mod,
 	enum sof_ipc_frame source_format;
 	enum sof_ipc_frame sink_format;
 	size_t data_size;
-	uint32_t sink_period_bytes;
 	int ret;
 
 	comp_info(dev, "entry");
@@ -178,15 +177,7 @@ static int mfcc_prepare(struct processing_module *mod,
 
 	/* get sink data format and period bytes */
 	sink_format = audio_stream_get_frm_fmt(&sinkb->stream);
-	sink_period_bytes = audio_stream_period_bytes(&sinkb->stream, dev->frames);
-	comp_info(dev, "source_format = %d, sink_format = %d",
-		  source_format, sink_format);
-	if (audio_stream_get_size(&sinkb->stream) < sink_period_bytes) {
-		comp_err(dev, "sink buffer size %d is insufficient < %d",
-			 audio_stream_get_size(&sinkb->stream), sink_period_bytes);
-		ret = -ENOMEM;
-		goto err;
-	}
+	comp_info(dev, "source_format = %d, sink_format = %d", source_format, sink_format);
 
 	cd->config = comp_get_data_blob(cd->model_handler, &data_size, NULL);
 

src/audio/mfcc/mfcc_common.c

@@ -45,13 +45,13 @@ static int mfcc_stft_process(const struct comp_dev *dev, struct mfcc_comp_data *
 	struct mfcc_fft *fft = &state->fft;
 	int mel_scale_shift;
 	int input_shift;
-	int i;
+	int i, j;
 	int m;
 	int cc_count = 0;
-	int32_t s;
-	int16_t mel_value;
-	int16_t peak;
-	int16_t clamp_value;
+	int64_t s;
+	int32_t mel_value;
+	int32_t peak;
+	int32_t clamp_value;
 
 	/* Phase 1, wait until whole fft_size is filled with valid data. This way
 	 * first output cepstral coefficients originate from streamed data and not
@@ -110,8 +110,8 @@ static int mfcc_stft_process(const struct comp_dev *dev, struct mfcc_comp_data *
 		fft_execute_32(fft->fft_plan, false);
 #endif
 
-		/* Convert powerspectrum to Mel band logarithmic spectrum */
-		mat_init_16b(state->mel_spectra, 1, state->dct.num_in, 7); /* Q8.7 */
+		/* Convert powerspectrum to Mel band logarithmic spectrum Q9.23 */
+		mat_init_16b(state->mel_spectra, 1, state->dct.num_in, 7); /* Q9.7 */
 
 		/* Compensate FFT lib scaling to Mel log values, e.g. for 512 long FFT
 		 * the fft_plan->len is 9. The scaling is 1/512. Subtract from input_shift it
@@ -121,54 +121,69 @@ static int mfcc_stft_process(const struct comp_dev *dev, struct mfcc_comp_data *
 #if MFCC_FFT_BITS == 16
 		psy_apply_mel_filterbank_16(&state->melfb, fft->fft_out, state->power_spectra,
 					    state->mel_spectra->data, mel_scale_shift);
+		/* Convert Q9.7 int16_t mel log to Q9.23 int32_t for downstream processing */
+		for (j = 0; j < state->dct.num_in; j++)
+			state->mel_log_32[j] = (int32_t)state->mel_spectra->data[j] << 16;
 #else
 		psy_apply_mel_filterbank_32(&state->melfb, fft->fft_out, state->power_spectra,
-					    state->mel_spectra->data, mel_scale_shift);
+					    state->mel_log_32, mel_scale_shift);
 #endif
 
 		if (state->mel_only) {
 			/* In Mel-only mode output Mel log spectra directly */
 			cc_count += state->dct.num_in;
 
-			/* Find peak mel value and track state->mmax */
+			/* Find peak mel value and track state->mmax in Q9.23 */
 			if (config->dynamic_mmax) {
-				peak = state->mel_spectra->data[0];
-				for (i = 1; i < state->dct.num_in; i++) {
-					if (state->mel_spectra->data[i] > peak)
-						peak = state->mel_spectra->data[i];
+				peak = state->mel_log_32[0];
+				for (j = 1; j < state->dct.num_in; j++) {
+					if (state->mel_log_32[j] > peak)
+						peak = state->mel_log_32[j];
 				}
 
 				/* Jump to peak immediately if higher, decay otherwise */
 				if (peak > state->mmax) {
 					state->mmax = peak;
 				} else {
-					/* Q8.7 * Q1.15, result Q8.7. The coefficient is small so
-					 * no need for saturation.
+					/* Q9.23 * Q1.15, result Q9.23. The coefficient is small
+					 * so no need for saturation.
 					 */
-					s = (int32_t)peak - state->mmax;
+					s = (int64_t)peak - state->mmax;
 					state->mmax +=
-						Q_MULTSR_32X32(s, config->mmax_coef, 7, 15, 7);
+						Q_MULTSR_32X32(s, config->mmax_coef, 23, 15, 23);
 				}
 			}
 
-			/* Clamp Mel values lower than mmax - top_db, add offset, and scale */
-			clamp_value = state->mmax - config->top_db;
-			for (i = 0; i < state->dct.num_in; i++) {
-				mel_value = state->mel_spectra->data[i];
+			/* Clamp Mel values lower than mmax - top_db, add offset, and scale.
+			 * Config top_db and mel_offset are Q9.7, shift to Q9.23.
+			 */
+			clamp_value = state->mmax - ((int32_t)config->top_db << 16);
+			for (j = 0; j < state->dct.num_in; j++) {
+				mel_value = state->mel_log_32[j];
 				if (mel_value < clamp_value)
 					mel_value = clamp_value;
 
-				/* Q8.7 * Q4.12, result 8.7 */
-				s = (int32_t)mel_value + config->mel_offset;
-				state->mel_spectra->data[i] =
-					sat_int16(Q_MULTSR_32X32(s, config->mel_scale, 7, 12, 7));
+				/* Q9.23 * Q4.12, result Q9.23 */
+				s = (int64_t)mel_value + ((int32_t)config->mel_offset << 16);
+				state->mel_log_32[j] =
+					sat_int32(Q_MULTSR_32X32(s, config->mel_scale, 23, 12, 23));
 			}
 
+			/* Store Q9.7 version in mel_spectra for s16 output mode */
+			for (j = 0; j < state->dct.num_in; j++)
+				state->mel_spectra->data[j] =
+					sat_int16(state->mel_log_32[j] >> 16);
+
 			/* Enable this to check mmax decay */
 			comp_dbg(dev, "state->mmax = %d", state->mmax);
 		} else {
+			/* Convert Q9.23 to Q9.7 for 16-bit DCT */
+			for (j = 0; j < state->dct.num_in; j++)
+				state->mel_spectra->data[j] =
+					sat_int16(state->mel_log_32[j] >> 16);
+
 			/* 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_init_16b(state->cepstral_coef, 1, state->dct.num_out, 7); /* Q9.7 */
 			mat_multiply(state->mel_spectra, state->dct.matrix, state->cepstral_coef);
 
 			/* Apply cepstral lifter */
@@ -189,6 +204,62 @@ static int mfcc_stft_process(const struct comp_dev *dev, struct mfcc_comp_data *
 	return cc_count;
 }
 
+void mfcc_fill_fft_buffer(struct mfcc_state *state)
+{
+	struct mfcc_buffer *buf = &state->buf;
+	struct mfcc_fft *fft = &state->fft;
+#if MFCC_FFT_BITS == 16
+	int16_t *d = &fft->fft_buf[fft->fft_fill_start_idx].real;
+	const int fft_elem_inc = sizeof(fft->fft_buf[0]) / sizeof(int16_t);
+#else
+	int32_t *d = &fft->fft_buf[fft->fft_fill_start_idx].real;
+	const int fft_elem_inc = sizeof(fft->fft_buf[0]) / sizeof(int32_t);
+#endif
+	int16_t *prev = state->prev_data;
+	int16_t *prev_end = prev + state->prev_data_size;
+	int16_t *r = buf->r_ptr;
+	int copied;
+	int nmax;
+	int n;
+	int j;
+
+	/* Copy overlapped samples from state buffer. The fft_buf has been
+	 * cleared by caller so imaginary part remains zero.
+	 */
+	while (prev < prev_end) {
+		*d = *prev++;
+		d += fft_elem_inc;
+	}
+
+	/* Copy hop size of new data from circular buffer */
+	for (copied = 0; copied < fft->fft_hop_size; copied += n) {
+		nmax = fft->fft_hop_size - copied;
+		n = mfcc_buffer_samples_without_wrap(buf, r);
+		n = MIN(n, nmax);
+		for (j = 0; j < n; j++) {
+			*d = *r++;
+			d += fft_elem_inc;
+		}
+		r = mfcc_buffer_wrap(buf, r);
+	}
+
+	buf->s_avail -= copied;
+	buf->s_free += copied;
+	buf->r_ptr = r;
+
+	/* Copy for next time data back to overlap buffer */
+#if MFCC_FFT_BITS == 16
+	d = (int16_t *)&fft->fft_buf[fft->fft_fill_start_idx + fft->fft_hop_size].real;
+#else
+	d = (int32_t *)&fft->fft_buf[fft->fft_fill_start_idx + fft->fft_hop_size].real;
+#endif
+	prev = state->prev_data;
+	while (prev < prev_end) {
+		*prev++ = *d;
+		d += fft_elem_inc;
+	}
+}
+
 #if CONFIG_FORMAT_S16LE
 static int16_t *mfcc_sink_copy_zero_s16(const struct audio_stream *sink, int16_t *w_ptr,
 					int samples)
@@ -338,6 +409,7 @@ void mfcc_s24_default(struct processing_module *mod, struct input_stream_buffer
 	int sink_samples;
 	int remain_s32;
 	int to_copy;
+	int k;
 
 	/* Get samples from source buffer */
 	mfcc_source_copy_s24(bsource, buf, &state->emph, frames, state->source_channel);
@@ -347,10 +419,15 @@ void mfcc_s24_default(struct processing_module *mod, struct input_stream_buffer
 
 	/* If new output produced, set up pointer into scratch data */
 	if (num_ceps > 0) {
-		if (state->mel_only)
-			state->out_data_ptr = state->mel_spectra->data;
-		else
+		if (state->mel_only) {
+			/* Convert mel_log_32 from Q9.23 to Q9.15 in-place */
+			for (k = 0; k < num_ceps; k++)
+				state->mel_log_32[k] >>= 8;
+
+			state->out_data_ptr_32 = state->mel_log_32;
+		} else {
 			state->out_data_ptr = state->cepstral_coef->data;
+		}
 
 		state->out_remain = num_ceps;
 		state->magic_pending = true;
@@ -366,18 +443,30 @@ void mfcc_s24_default(struct processing_module *mod, struct input_stream_buffer
 		state->magic_pending = false;
 	}
 
-	/* Write cepstral/mel data packed as int32_t from scratch buffer */
-	remain_s32 = (state->out_remain + 1) / 2;
-	to_copy = MIN(remain_s32, sink_samples);
-	if (to_copy > 0) {
-		w_ptr = mfcc_sink_copy_data_s32(sink, w_ptr, to_copy,
-						(int32_t *)state->out_data_ptr);
-		state->out_data_ptr += to_copy * 2;
-		state->out_remain -= to_copy * 2;
-		if (state->out_remain < 0)
-			state->out_remain = 0;
-
-		sink_samples -= to_copy;
+	if (state->mel_only) {
+		/* Write 32-bit mel data Q9.15, one value per int32_t */
+		to_copy = MIN(state->out_remain, sink_samples);
+		if (to_copy > 0) {
+			w_ptr = mfcc_sink_copy_data_s32(sink, w_ptr, to_copy,
+							state->out_data_ptr_32);
+			state->out_data_ptr_32 += to_copy;
+			state->out_remain -= to_copy;
+			sink_samples -= to_copy;
+		}
+	} else {
+		/* Write cepstral data packed as int32_t from scratch buffer */
+		remain_s32 = (state->out_remain + 1) / 2;
+		to_copy = MIN(remain_s32, sink_samples);
+		if (to_copy > 0) {
+			w_ptr = mfcc_sink_copy_data_s32(sink, w_ptr, to_copy,
+							(int32_t *)state->out_data_ptr);
+			state->out_data_ptr += to_copy * 2;
+			state->out_remain -= to_copy * 2;
+			if (state->out_remain < 0)
+				state->out_remain = 0;
+
+			sink_samples -= to_copy;
+		}
 	}
 
 	/* Zero-fill remaining sink samples */
@@ -409,10 +498,11 @@ void mfcc_s32_default(struct processing_module *mod, struct input_stream_buffer
 
 	/* If new output produced, set up pointer into scratch data */
 	if (num_ceps > 0) {
-		if (state->mel_only)
-			state->out_data_ptr = state->mel_spectra->data;
-		else
+		if (state->mel_only) {
+			state->out_data_ptr_32 = state->mel_log_32;
+		} else {
 			state->out_data_ptr = state->cepstral_coef->data;
+		}
 
 		state->out_remain = num_ceps;
 		state->magic_pending = true;
@@ -428,18 +518,30 @@ void mfcc_s32_default(struct processing_module *mod, struct input_stream_buffer
 		state->magic_pending = false;
 	}
 
-	/* Write cepstral/mel data packed as int32_t from scratch buffer */
-	remain_s32 = (state->out_remain + 1) / 2;
-	to_copy = MIN(remain_s32, sink_samples);
-	if (to_copy > 0) {
-		w_ptr = mfcc_sink_copy_data_s32(sink, w_ptr, to_copy,
-						(int32_t *)state->out_data_ptr);
-		state->out_data_ptr += to_copy * 2;
-		state->out_remain -= to_copy * 2;
-		if (state->out_remain < 0)
-			state->out_remain = 0;
-
-		sink_samples -= to_copy;
+	if (state->mel_only) {
+		/* Write 32-bit mel data Q9.23, one value per int32_t */
+		to_copy = MIN(state->out_remain, sink_samples);
+		if (to_copy > 0) {
+			w_ptr = mfcc_sink_copy_data_s32(sink, w_ptr, to_copy,
+							state->out_data_ptr_32);
+			state->out_data_ptr_32 += to_copy;
+			state->out_remain -= to_copy;
+			sink_samples -= to_copy;
+		}
+	} else {
+		/* Write cepstral data packed as int32_t from scratch buffer */
+		remain_s32 = (state->out_remain + 1) / 2;
+		to_copy = MIN(remain_s32, sink_samples);
+		if (to_copy > 0) {
+			w_ptr = mfcc_sink_copy_data_s32(sink, w_ptr, to_copy,
+							(int32_t *)state->out_data_ptr);
+			state->out_data_ptr += to_copy * 2;
+			state->out_remain -= to_copy * 2;
+			if (state->out_remain < 0)
+				state->out_remain = 0;
+
+			sink_samples -= to_copy;
+		}
 	}
 
 	/* Zero-fill remaining sink samples */