EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add C++ runtime for Tele-AI/TeleSpeech-ASR (#970)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-06-05 00:26:40 +08:00
committed by GitHub
parent f8dbc10146
commit fd5a0d1e00
52 changed files with 1052 additions and 145 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
wasm/asr/sherpa-onnx-asr.js
View File

@@ -529,7 +529,7 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
const tdnn = initSherpaOnnxOfflineTdnnModelConfig(config.tdnn, Module);
const len = transducer.len + paraformer.len + nemoCtc.len + whisper.len +
tdnn.len + 7 * 4;
tdnn.len + 8 * 4;
const ptr = Module._malloc(len);
let offset = 0;
@@ -553,9 +553,11 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
const modelTypeLen = Module.lengthBytesUTF8(config.modelType) + 1;
const modelingUnitLen = Module.lengthBytesUTF8(config.modelingUnit || '') + 1;
const bpeVocabLen = Module.lengthBytesUTF8(config.bpeVocab || '') + 1;
const teleSpeechCtcLen =
Module.lengthBytesUTF8(config.teleSpeechCtc || '') + 1;
const bufferLen =
tokensLen + providerLen + modelTypeLen + modelingUnitLen + bpeVocabLen;
const bufferLen = tokensLen + providerLen + modelTypeLen + modelingUnitLen +
bpeVocabLen + teleSpeechCtcLen;
const buffer = Module._malloc(bufferLen);
offset = 0;
@@ -575,6 +577,10 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
Module.stringToUTF8(config.bpeVocab || '', buffer + offset, bpeVocabLen);
offset += bpeVocabLen;
Module.stringToUTF8(
config.teleSpeechCtc || '', buffer + offset, teleSpeechCtcLen);
offset += teleSpeechCtcLen;
offset =
transducer.len + paraformer.len + nemoCtc.len + whisper.len + tdnn.len;
Module.setValue(ptr + offset, buffer, 'i8*'); // tokens
@@ -604,6 +610,13 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
'i8*'); // bpeVocab
offset += 4;
Module.setValue(
ptr + offset,
buffer + tokensLen + providerLen + modelTypeLen + modelingUnitLen +
bpeVocabLen,
'i8*'); // teleSpeechCtc
offset += 4;
return {
buffer: buffer, ptr: ptr, len: len, transducer: transducer,
paraformer: paraformer, nemoCtc: nemoCtc, whisper: whisper, tdnn: tdnn

3
wasm/nodejs/sherpa-onnx-wasm-nodejs.cc
View File

@@ -23,7 +23,7 @@ static_assert(sizeof(SherpaOnnxOfflineModelConfig) ==
sizeof(SherpaOnnxOfflineParaformerModelConfig) +
sizeof(SherpaOnnxOfflineNemoEncDecCtcModelConfig) +
sizeof(SherpaOnnxOfflineWhisperModelConfig) +
sizeof(SherpaOnnxOfflineTdnnModelConfig) + 7 * 4,
sizeof(SherpaOnnxOfflineTdnnModelConfig) + 8 * 4,
"");
static_assert(sizeof(SherpaOnnxFeatureConfig) == 2 * 4, "");
static_assert(sizeof(SherpaOnnxOfflineRecognizerConfig) ==
@@ -92,6 +92,7 @@ void PrintOfflineRecognizerConfig(SherpaOnnxOfflineRecognizerConfig *config) {
fprintf(stdout, "model type: %s\n", model_config->model_type);
fprintf(stdout, "modeling unit: %s\n", model_config->modeling_unit);
fprintf(stdout, "bpe vocab: %s\n", model_config->bpe_vocab);
fprintf(stdout, "telespeech_ctc: %s\n", model_config->telespeech_ctc);
fprintf(stdout, "----------feat config----------\n");
fprintf(stdout, "sample rate: %d\n", feat->sample_rate);