pinebuds/apps/audioplayers/bt_sco_chain_thirdparty.c

160 lines
4.3 KiB
C
Raw Permalink Normal View History

#include "audio_dump.h"
2022-08-15 04:20:27 -05:00
#include "bt_sco_chain.h"
#include "hal_trace.h"
#include "speech_cfg.h"
#include "speech_memory.h"
#include "speech_utils.h"
2022-08-15 04:20:27 -05:00
#if defined(SPEECH_TX_24BIT)
int32_t *aec_echo_buf = NULL;
// Use to free buffer
static int32_t *aec_echo_buf_ptr = NULL;
#else
short *aec_echo_buf = NULL;
// Use to free buffer
static short *aec_echo_buf_ptr;
#endif
#if defined(MSBC_8K_SAMPLE_RATE)
#define SPEECH_HEAP_RESERVE_SIZE (1024 * 4)
#else
#define SPEECH_HEAP_RESERVE_SIZE (1024 * 4)
#endif
#if defined(SPEECH_TX_DC_FILTER)
SpeechDcFilterState *speech_tx_dc_filter_st = NULL;
#endif
int speech_init(int tx_sample_rate, int rx_sample_rate, int tx_frame_ms,
int rx_frame_ms, int sco_frame_ms, uint8_t *buf, int len) {
// we shoule keep a minmum buffer for speech heap
// MSBC_16K_SAMPLE_RATE = 0, 560 bytes
// MSBC_16K_SAMPLE_RATE = 1, 2568 bytes
speech_heap_init(buf, SPEECH_HEAP_RESERVE_SIZE);
2022-08-15 04:20:27 -05:00
uint8_t *free_buf = buf + SPEECH_HEAP_RESERVE_SIZE;
int free_len = len - SPEECH_HEAP_RESERVE_SIZE;
2022-08-15 04:20:27 -05:00
// use free_buf for your algorithm
memset(free_buf, 0, free_len);
2022-08-15 04:20:27 -05:00
int frame_len = SPEECH_FRAME_MS_TO_LEN(tx_sample_rate, tx_frame_ms);
2022-08-15 04:20:27 -05:00
#if defined(SPEECH_TX_24BIT)
aec_echo_buf = (int32_t *)speech_calloc(frame_len, sizeof(int32_t));
2022-08-15 04:20:27 -05:00
#else
aec_echo_buf = (short *)speech_calloc(frame_len, sizeof(short));
2022-08-15 04:20:27 -05:00
#endif
aec_echo_buf_ptr = aec_echo_buf;
2022-08-15 04:20:27 -05:00
#if defined(SPEECH_TX_DC_FILTER)
int channel_num = SPEECH_CODEC_CAPTURE_CHANNEL_NUM;
int data_separation = 0;
SpeechDcFilterConfig dc_filter_cfg = {
.bypass = 0,
.gain = 0.f,
};
speech_tx_dc_filter_st =
speech_dc_filter_create(tx_sample_rate, frame_len, &dc_filter_cfg);
speech_dc_filter_ctl(speech_tx_dc_filter_st, SPEECH_DC_FILTER_SET_CHANNEL_NUM,
&channel_num);
speech_dc_filter_ctl(speech_tx_dc_filter_st,
SPEECH_DC_FILTER_SET_DATA_SEPARATION, &data_separation);
2022-08-15 04:20:27 -05:00
#endif
audio_dump_init(frame_len, sizeof(int16_t), 3);
2022-08-15 04:20:27 -05:00
return 0;
2022-08-15 04:20:27 -05:00
}
int speech_deinit(void) {
2022-08-15 04:20:27 -05:00
#if defined(SPEECH_TX_DC_FILTER)
speech_dc_filter_destroy(speech_tx_dc_filter_st);
2022-08-15 04:20:27 -05:00
#endif
speech_free(aec_echo_buf_ptr);
2022-08-15 04:20:27 -05:00
size_t total = 0, used = 0, max_used = 0;
speech_memory_info(&total, &used, &max_used);
TRACE(3, "SPEECH MALLOC MEM: total - %d, used - %d, max_used - %d.", total,
used, max_used);
ASSERT(used == 0, "[%s] used != 0", __func__);
2022-08-15 04:20:27 -05:00
return 0;
2022-08-15 04:20:27 -05:00
}
#if defined(BONE_SENSOR_TDM)
extern void bt_sco_get_tdm_buffer(uint8_t **buf, uint32_t *len);
#endif
int speech_tx_process(void *_pcm_buf, void *_ref_buf, int *_pcm_len) {
2022-08-15 04:20:27 -05:00
#if defined(SPEECH_TX_24BIT)
int32_t *pcm_buf = (int32_t *)_pcm_buf;
int32_t *ref_buf = (int32_t *)_ref_buf;
2022-08-15 04:20:27 -05:00
#else
int16_t *pcm_buf = (int16_t *)_pcm_buf;
int16_t *ref_buf = (int16_t *)_ref_buf;
2022-08-15 04:20:27 -05:00
#endif
int pcm_len = *_pcm_len;
2022-08-15 04:20:27 -05:00
#if defined(BONE_SENSOR_TDM)
uint8_t *bone_buf = NULL;
uint32_t bone_len = 0;
bt_sco_get_tdm_buffer(&bone_buf, &bone_len);
2022-08-15 04:20:27 -05:00
#endif
audio_dump_clear_up();
audio_dump_add_channel_data(0, ref_buf,
pcm_len / SPEECH_CODEC_CAPTURE_CHANNEL_NUM);
audio_dump_add_channel_data_from_multi_channels(
1, pcm_buf, pcm_len / SPEECH_CODEC_CAPTURE_CHANNEL_NUM,
SPEECH_CODEC_CAPTURE_CHANNEL_NUM, 0);
audio_dump_add_channel_data_from_multi_channels(
2, pcm_buf, pcm_len / SPEECH_CODEC_CAPTURE_CHANNEL_NUM,
SPEECH_CODEC_CAPTURE_CHANNEL_NUM, 1);
2022-08-15 04:20:27 -05:00
#if defined(BONE_SENSOR_TDM)
audio_dump_add_channel_data(3, bone_buf,
pcm16_len / SPEECH_CODEC_CAPTURE_CHANNEL_NUM);
2022-08-15 04:20:27 -05:00
#endif
audio_dump_run();
2022-08-15 04:20:27 -05:00
#if defined(SPEECH_TX_DC_FILTER)
#if defined(SPEECH_TX_24BIT)
speech_dc_filter_process_int24(speech_tx_dc_filter_st, pcm_buf, pcm_len);
2022-08-15 04:20:27 -05:00
#else
speech_dc_filter_process(speech_tx_dc_filter_st, pcm_buf, pcm_len);
2022-08-15 04:20:27 -05:00
#endif
#endif
// Add your algrithm here and disable #if macro
2022-08-15 04:20:27 -05:00
#if 1
for (int i = 0, j = 0; i < pcm_len;
i += SPEECH_CODEC_CAPTURE_CHANNEL_NUM, j++) {
// choose main microphone data
pcm_buf[j] = pcm_buf[i];
// choose reference data, i.e. loopback
// pcm16_buf[j] = ref16_buf[j];
}
pcm_len /= SPEECH_CODEC_CAPTURE_CHANNEL_NUM;
2022-08-15 04:20:27 -05:00
#endif
#if defined(BONE_SENSOR_TDM)
memcpy(pcm_buf, bone_buf, bone_len);
2022-08-15 04:20:27 -05:00
#endif
*_pcm_len = pcm_len;
2022-08-15 04:20:27 -05:00
return 0;
2022-08-15 04:20:27 -05:00
}
int speech_rx_process(void *pcm_buf, int *pcm_len) {
// Add your algorithm here
return 0;
2022-08-15 04:20:27 -05:00
}