extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-04-16 23:38:45 +02:00
Code Issues Packages Projects Releases Wiki Activity

whisper : fixed Beam Search Strategy and exposed whisper_pcm_to_mel_phase_vocoder (#474)

Browse Source 
Co-authored-by: Sandro Hanea <sandrohanea@microsoft.com>
This commit is contained in:
sandrohanea 2023-02-08 08:01:47 +01:00 committed by GitHub
parent 4dd7119deb
commit 2bfe0ebc0f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
whisper.cpp
View File

@ -2905,7 +2905,7 @@ const char * whisper_print_system_info(void) {
struct whisper_full_params whisper_full_default_params(enum whisper_sampling_strategy strategy) { struct whisper_full_params whisper_full_default_params(enum whisper_sampling_strategy strategy) {
struct whisper_full_params result = { struct whisper_full_params result = {
/*.strategy =*/ WHISPER_SAMPLING_GREEDY, /*.strategy =*/ strategy,
/*.n_threads =*/ std::min(4, (int32_t) std::thread::hardware_concurrency()), /*.n_threads =*/ std::min(4, (int32_t) std::thread::hardware_concurrency()),
/*.n_max_text_ctx =*/ 16384, /*.n_max_text_ctx =*/ 16384,
@ -3829,7 +3829,7 @@ int whisper_full(
auto & cur = beam_candidates[cur_c++]; auto & cur = beam_candidates[cur_c++];
while (beam_candidates[cur_c].sequence.sum_logprobs_all == cur.sequence.sum_logprobs_all && i > 0) { while (beam_candidates.size() > cur_c && beam_candidates[cur_c].sequence.sum_logprobs_all == cur.sequence.sum_logprobs_all && i > 0) {
++cur_c; ++cur_c;
} }

10
whisper.h
View File

@ -113,6 +113,16 @@ extern "C" {
int n_samples, int n_samples,
int n_threads); int n_threads);
// Convert RAW PCM audio to log mel spectrogram but applies a Phase Vocoder to speed up the audio x2.
// The resulting spectrogram is stored inside the provided whisper context.
// Returns 0 on success
WHISPER_API int whisper_pcm_to_mel_phase_vocoder(
struct whisper_context* ctx,
const float* samples,
int n_samples,
int n_threads);
// This can be used to set a custom log mel spectrogram inside the provided whisper context. // This can be used to set a custom log mel spectrogram inside the provided whisper context.
// Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram. // Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
// n_mel must be 80 // n_mel must be 80