public_git
/
whisper.cpp
mirror of https://github.com/ggml-org/whisper.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

examples : remove audio_ctx param from parakeet-server.cpp [no ci] 

The motivation for this is that setting audio_ctx smaller than the
model default forces the encoder to process shorter chunks than it was
trained on which will hurt accuracy.

In the future a streaming model should be able to handles this instead.
This commit is contained in:
Daniel Bevenius 2026-06-25 11:45:23 +02:00
parent 9b2164b4d5
commit 5f306f6acc
No known key found for this signature in database
1 changed files with 0 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
examples/parakeet-server/parakeet-server.cpp
View File

@ -24,8 +24,6 @@ struct parakeet_params {
int32_t offset_ms = 0;
int32_t duration_ms = 0;
int32_t audio_ctx = 0; // 0 = use default
bool use_gpu = true;
int32_t gpu_device = 0;
@ -42,7 +40,6 @@ static void parakeet_print_usage(int /*argc*/, char ** argv, const parakeet_para
fprintf(stderr, " -t N, --threads N [%-7d] number of threads to use during computation\n", params.n_threads);
fprintf(stderr, " -ot N, --offset-t N [%-7d] time offset in milliseconds\n", params.offset_ms);
fprintf(stderr, " -d N, --duration N [%-7d] duration of audio to process in milliseconds\n", params.duration_ms);
fprintf(stderr, " -ac N, --audio-ctx N [%-7d] audio context size (0 = use default)\n", params.audio_ctx);
fprintf(stderr, " -m FNAME, --model FNAME [%-7s] model path\n", params.model.c_str());
fprintf(stderr, " -ng, --no-gpu [%-7s] do not use GPU\n", params.use_gpu ? "false" : "true");
fprintf(stderr, " -dev N, --device N [%-7d] GPU device ID\n", params.gpu_device);
@ -72,7 +69,6 @@ static bool parakeet_params_parse(int argc, char ** argv, parakeet_params & para
else if (arg == "-t" || arg == "--threads") { params.n_threads = std::stoi(argv[++i]); }
else if (arg == "-ot" || arg == "--offset-t") { params.offset_ms = std::stoi(argv[++i]); }
else if (arg == "-d" || arg == "--duration") { params.duration_ms = std::stoi(argv[++i]); }
else if (arg == "-ac" || arg == "--audio-ctx") { params.audio_ctx = std::stoi(argv[++i]); }
else if (arg == "-m" || arg == "--model") { params.model = argv[++i]; }
else if (arg == "-ng" || arg == "--no-gpu") { params.use_gpu = false; }
else if (arg == "-dev" || arg == "--device") { params.gpu_device = std::stoi(argv[++i]); }
@ -100,9 +96,6 @@ static void get_req_parameters(const Request & req, parakeet_params & params) {
if (req.has_file("duration")) {
params.duration_ms = std::stoi(req.get_file_value("duration").content);
}
if (req.has_file("audio_ctx")) {
params.audio_ctx = std::stoi(req.get_file_value("audio_ctx").content);
}
if (req.has_file("response_format")) {
params.response_format = req.get_file_value("response_format").content;
}
@ -304,7 +297,6 @@ int main(int argc, char ** argv) {
fparams.offset_ms = cur_params.offset_ms;
fparams.duration_ms = cur_params.duration_ms;
fparams.no_context = true;
fparams.audio_ctx = cur_params.audio_ctx;
// Abort callback for HTTP disconnect
fparams.abort_callback = [](void * user_data) {