how to translate realtime pcm data to text? about go-astideepspeech

[asticode]

Your detectVoice function is never called. Is that normal ?Your detectVoice function is never called. Is that normal ?

[jackleibest]

actually it was called by other function whenever the pcm data decoded by ffmpeg

[asticode]

Can you paste a fully working example that I can run myself ?

[jackleibest]

my code is working on a stream server with ffmpeg sdk, so it can't working stand alone, can you show us an example for live stream translate?

[asticode]

I've never used the live stream portion of DeepSpeech so I'd need some time to work out an example.

Before that, can you confirm that audio data you're feeding DeepSpeech is 16 bits / 16kHz PCM samples ?

[jackleibest]

yeah, surely, just resample the audio to 16bits / 16kHz pcm data, here is the c lang code for aac to pcm:
static int aacdec_decode(aacdec_t *m, uint8_t *data, int len) {
AVPacket pkt;
av_init_packet(&pkt);
pkt.data = data;
pkt.size = len;
//av_log(m->ctx, AV_LOG_DEBUG, "decode %p\n", m);
int got;
int res = avcodec_decode_audio4(m->ctx, m->f, &got, &pkt);
if(got){
if(m->swr == NULL){
m->src_rate = m->ctx->sample_rate;
m->dst_rate = 16000;
m->swr = swr_alloc();
av_opt_set_int(m->swr, "in_channel_layout", m->ctx->channel_layout, 0);
av_opt_set_int(m->swr, "in_sample_rate", m->src_rate, 0);
av_opt_set_sample_fmt(m->swr, "in_sample_fmt", m->ctx->sample_fmt, 0);
av_opt_set_int(m->swr, "out_channel_layout", AV_CH_LAYOUT_MONO, 0);
av_opt_set_int(m->swr, "out_sample_rate", m->dst_rate, 0);
av_opt_set_sample_fmt(m->swr, "out_sample_fmt", AV_SAMPLE_FMT_S16, 0);
swr_init(m->swr);
}
int dst_linesize;
int dst_nb_channels = av_get_channel_layout_nb_channels(AV_CH_LAYOUT_MONO);
int src_nb_samples = m->f->nb_samples;
m->frameBuff = (uint8_t *)av_malloc(MAX_AUDIO_FRAME_SIZE*2);
int dst_nb_samples = av_rescale_rnd(swr_get_delay(m->swr, m->src_rate) + src_nb_samples, m->dst_rate, m->src_rate, AV_ROUND_UP);
av_samples_alloc_array_and_samples(m->frameBuff, &dst_linesize, dst_nb_channels,dst_nb_samples, AV_SAMPLE_FMT_S16, 0);
int len = swr_convert(m->swr, &m->frameBuff, dst_nb_samples, (const uint8_t **)m->f->data, src_nb_samples);
int total = av_samples_get_buffer_size(&dst_linesize, dst_nb_channels, len, AV_SAMPLE_FMT_S16, 1);
}
av_packet_unref(&pkt);
return res;
}

[asticode]

What do you get at fmt.Println("result: ", result) ?

Also, what happens if you save all samples to a .wav file and use this lib's cmd to translate it ? In other words, does the non-streaming translation work with your samples ?

[jackleibest]

just get the response: "result: ". I did not try out the non-streaming translation with my samples, because I failed to save the pcm to wav file successfully, but I can play the raw pcm with audacity or ffplay whenever save them to PCM file.

[asticode]

Thing is, we need to be sure your samples give correct results with the non-streaming part of this lib.

To save your samples to a wav file you can get inspiration from this bit of code.

Once it's done, let me know what results you get from the non-streaming translation.

[jackleibest]

ok, thanks, I will try it in someday, too busy on job things at present.~~