|
| 1 | +// c-api-examples/keywords-spotter-buffered-tokens-keywords-c-api.c |
| 2 | +// |
| 3 | +// Copyright (c) 2024 Xiaomi Corporation |
| 4 | +// Copyright (c) 2024 Luo Xiao |
| 5 | + |
| 6 | +// |
| 7 | +// This file demonstrates how to use keywords spotter with sherpa-onnx's C |
| 8 | +// API and with tokens and keywords loaded from buffered strings instead of from |
| 9 | +// external files API. |
| 10 | +// clang-format off |
| 11 | +// |
| 12 | +// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/kws-models/sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile.tar.bz2 |
| 13 | +// tar xvf sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile.tar.bz2 |
| 14 | +// rm sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile.tar.bz2 |
| 15 | +// |
| 16 | +// clang-format on |
| 17 | + |
| 18 | +#include <stdio.h> |
| 19 | +#include <stdlib.h> |
| 20 | +#include <string.h> |
| 21 | + |
| 22 | +#include "sherpa-onnx/c-api/c-api.h" |
| 23 | + |
| 24 | +static size_t ReadFile(const char *filename, const char **buffer_out) { |
| 25 | + FILE *file = fopen(filename, "r"); |
| 26 | + if (file == NULL) { |
| 27 | + fprintf(stderr, "Failed to open %s\n", filename); |
| 28 | + return -1; |
| 29 | + } |
| 30 | + fseek(file, 0L, SEEK_END); |
| 31 | + long size = ftell(file); |
| 32 | + rewind(file); |
| 33 | + *buffer_out = malloc(size); |
| 34 | + if (*buffer_out == NULL) { |
| 35 | + fclose(file); |
| 36 | + fprintf(stderr, "Memory error\n"); |
| 37 | + return -1; |
| 38 | + } |
| 39 | + size_t read_bytes = fread(*buffer_out, 1, size, file); |
| 40 | + if (read_bytes != size) { |
| 41 | + printf("Errors occured in reading the file %s\n", filename); |
| 42 | + free((void *)*buffer_out); |
| 43 | + *buffer_out = NULL; |
| 44 | + fclose(file); |
| 45 | + return -1; |
| 46 | + } |
| 47 | + fclose(file); |
| 48 | + return read_bytes; |
| 49 | +} |
| 50 | + |
| 51 | +int32_t main() { |
| 52 | + const char *wav_filename = |
| 53 | + "sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/test_wavs/" |
| 54 | + "6.wav"; |
| 55 | + const char *encoder_filename = |
| 56 | + "sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/" |
| 57 | + "encoder-epoch-12-avg-2-chunk-16-left-64.int8.onnx"; |
| 58 | + const char *decoder_filename = |
| 59 | + "sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/" |
| 60 | + "decoder-epoch-12-avg-2-chunk-16-left-64.onnx"; |
| 61 | + const char *joiner_filename = |
| 62 | + "sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/" |
| 63 | + "joiner-epoch-12-avg-2-chunk-16-left-64.int8.onnx"; |
| 64 | + const char *provider = "cpu"; |
| 65 | + const char *tokens_filename = |
| 66 | + "sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/tokens.txt"; |
| 67 | + const char *keywords_filename = |
| 68 | + "sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/test_wavs/" |
| 69 | + "test_keywords.txt"; |
| 70 | + const SherpaOnnxWave *wave = SherpaOnnxReadWave(wav_filename); |
| 71 | + if (wave == NULL) { |
| 72 | + fprintf(stderr, "Failed to read %s\n", wav_filename); |
| 73 | + return -1; |
| 74 | + } |
| 75 | + |
| 76 | + // reading tokens and keywords to buffers |
| 77 | + const char *tokens_buf; |
| 78 | + size_t token_buf_size = ReadFile(tokens_filename, &tokens_buf); |
| 79 | + if (token_buf_size < 1) { |
| 80 | + fprintf(stderr, "Please check your tokens.txt!\n"); |
| 81 | + free((void *)tokens_buf); |
| 82 | + return -1; |
| 83 | + } |
| 84 | + const char *keywords_buf; |
| 85 | + size_t keywords_buf_size = ReadFile(keywords_filename, &keywords_buf); |
| 86 | + if (keywords_buf_size < 1) { |
| 87 | + fprintf(stderr, "Please check your keywords.txt!\n"); |
| 88 | + free((void *)keywords_buf); |
| 89 | + return -1; |
| 90 | + } |
| 91 | + |
| 92 | + // Zipformer config |
| 93 | + SherpaOnnxOnlineTransducerModelConfig zipformer_config; |
| 94 | + memset(&zipformer_config, 0, sizeof(zipformer_config)); |
| 95 | + zipformer_config.encoder = encoder_filename; |
| 96 | + zipformer_config.decoder = decoder_filename; |
| 97 | + zipformer_config.joiner = joiner_filename; |
| 98 | + |
| 99 | + // Online model config |
| 100 | + SherpaOnnxOnlineModelConfig online_model_config; |
| 101 | + memset(&online_model_config, 0, sizeof(online_model_config)); |
| 102 | + online_model_config.debug = 1; |
| 103 | + online_model_config.num_threads = 1; |
| 104 | + online_model_config.provider = provider; |
| 105 | + online_model_config.tokens_buf = tokens_buf; |
| 106 | + online_model_config.tokens_buf_size = token_buf_size; |
| 107 | + online_model_config.transducer = zipformer_config; |
| 108 | + |
| 109 | + // Keywords-spotter config |
| 110 | + SherpaOnnxKeywordSpotterConfig keywords_spotter_config; |
| 111 | + memset(&keywords_spotter_config, 0, sizeof(keywords_spotter_config)); |
| 112 | + keywords_spotter_config.max_active_paths = 4; |
| 113 | + keywords_spotter_config.keywords_threshold = 0.1; |
| 114 | + keywords_spotter_config.keywords_score = 3.0; |
| 115 | + keywords_spotter_config.model_config = online_model_config; |
| 116 | + keywords_spotter_config.keywords_buf = keywords_buf; |
| 117 | + keywords_spotter_config.keywords_buf_size = keywords_buf_size; |
| 118 | + |
| 119 | + SherpaOnnxKeywordSpotter *keywords_spotter = |
| 120 | + SherpaOnnxCreateKeywordSpotter(&keywords_spotter_config); |
| 121 | + |
| 122 | + free((void *)tokens_buf); |
| 123 | + tokens_buf = NULL; |
| 124 | + free((void *)keywords_buf); |
| 125 | + keywords_buf = NULL; |
| 126 | + |
| 127 | + if (keywords_spotter == NULL) { |
| 128 | + fprintf(stderr, "Please check your config!\n"); |
| 129 | + SherpaOnnxFreeWave(wave); |
| 130 | + return -1; |
| 131 | + } |
| 132 | + |
| 133 | + SherpaOnnxOnlineStream *stream = |
| 134 | + SherpaOnnxCreateKeywordStream(keywords_spotter); |
| 135 | + |
| 136 | + const SherpaOnnxDisplay *display = SherpaOnnxCreateDisplay(50); |
| 137 | + int32_t segment_id = 0; |
| 138 | + |
| 139 | +// simulate streaming. You can choose an arbitrary N |
| 140 | +#define N 3200 |
| 141 | + |
| 142 | + fprintf(stderr, "sample rate: %d, num samples: %d, duration: %.2f s\n", |
| 143 | + wave->sample_rate, wave->num_samples, |
| 144 | + (float)wave->num_samples / wave->sample_rate); |
| 145 | + |
| 146 | + int32_t k = 0; |
| 147 | + while (k < wave->num_samples) { |
| 148 | + int32_t start = k; |
| 149 | + int32_t end = |
| 150 | + (start + N > wave->num_samples) ? wave->num_samples : (start + N); |
| 151 | + k += N; |
| 152 | + |
| 153 | + SherpaOnnxOnlineStreamAcceptWaveform(stream, wave->sample_rate, |
| 154 | + wave->samples + start, end - start); |
| 155 | + while (SherpaOnnxIsKeywordStreamReady(keywords_spotter, stream)) { |
| 156 | + SherpaOnnxDecodeKeywordStream(keywords_spotter, stream); |
| 157 | + } |
| 158 | + |
| 159 | + const SherpaOnnxKeywordResult *r = |
| 160 | + SherpaOnnxGetKeywordResult(keywords_spotter, stream); |
| 161 | + |
| 162 | + if (strlen(r->keyword)) { |
| 163 | + SherpaOnnxPrint(display, segment_id, r->keyword); |
| 164 | + } |
| 165 | + |
| 166 | + SherpaOnnxDestroyKeywordResult(r); |
| 167 | + } |
| 168 | + |
| 169 | + // add some tail padding |
| 170 | + float tail_paddings[4800] = {0}; // 0.3 seconds at 16 kHz sample rate |
| 171 | + SherpaOnnxOnlineStreamAcceptWaveform(stream, wave->sample_rate, tail_paddings, |
| 172 | + 4800); |
| 173 | + |
| 174 | + SherpaOnnxFreeWave(wave); |
| 175 | + |
| 176 | + SherpaOnnxOnlineStreamInputFinished(stream); |
| 177 | + while (SherpaOnnxIsKeywordStreamReady(keywords_spotter, stream)) { |
| 178 | + SherpaOnnxDecodeKeywordStream(keywords_spotter, stream); |
| 179 | + } |
| 180 | + |
| 181 | + const SherpaOnnxKeywordResult *r = |
| 182 | + SherpaOnnxGetKeywordResult(keywords_spotter, stream); |
| 183 | + |
| 184 | + if (strlen(r->keyword)) { |
| 185 | + SherpaOnnxPrint(display, segment_id, r->keyword); |
| 186 | + } |
| 187 | + |
| 188 | + SherpaOnnxDestroyKeywordResult(r); |
| 189 | + |
| 190 | + SherpaOnnxDestroyDisplay(display); |
| 191 | + SherpaOnnxDestroyOnlineStream(stream); |
| 192 | + SherpaOnnxDestroyKeywordSpotter(keywords_spotter); |
| 193 | + fprintf(stderr, "\n"); |
| 194 | + |
| 195 | + return 0; |
| 196 | +} |
0 commit comments