-cJj=$nddlmZddlmZmZddlmZmZmZm Z m Z m Z ddlm Z defdZGddeZ y ) )Path)ListOptional)FeatureExtractorConfigKeywordSpotterConfigOnlineModelConfigOnlineTransducerModelConfig OnlineStreamProviderConfig)KeywordSpotterfcJt|js J|dy)Nz does not exist)ris_file)r s e/Users/ahmed/devFolder/claude-voice/.venv/lib/python3.12/site-packages/sherpa_onnx/keyword_spotter.py_assert_file_existsrs" 7?? 3?33 ceZdZdZ ddededededededed ed ed ed ed ededefdZdefdZ dde efdZ defdZ de efdZdedefdZdedefdZdede efdZdede efdZy) r uB A class for keyword spotting. It uses streaming transducer models with keyword lists. Example using pre-defined keywords:: import numpy as np import sherpa_onnx import soundfile as sf kws = sherpa_onnx.KeywordSpotter( tokens="./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/tokens.txt", encoder="./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/encoder-epoch-12-avg-2-chunk-16-left-64.int8.onnx", decoder="./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/decoder-epoch-12-avg-2-chunk-16-left-64.onnx", joiner="./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/joiner-epoch-12-avg-2-chunk-16-left-64.int8.onnx", keywords_file="./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01-mobile/test_wavs/test_keywords.txt", num_threads=2, provider="cpu", ) audio, sample_rate = sf.read("test.wav", dtype="float32") tail_paddings = np.zeros(int(0.66 * sample_rate), dtype=np.float32) stream = kws.create_stream() stream.accept_waveform(sample_rate, audio) stream.accept_waveform(sample_rate, tail_paddings) stream.input_finished() while kws.is_ready(stream): kws.decode_stream(stream) r = kws.get_result(stream) if r != "": # Remember to call reset right after detecting a keyword kws.reset_stream(stream) print(f"Detected: {r}") Example with inline keywords:: # Add extra keywords at stream creation time stream = kws.create_stream("y ǎn y uán @演员/zh ī m íng @知名") stream.accept_waveform(sample_rate, audio) stream.accept_waveform(sample_rate, tail_paddings) stream.input_finished() while kws.is_ready(stream): kws.decode_stream(stream) r = kws.get_result(stream) if r != "": kws.reset_stream(stream) print(f"Detected: {r}") Please refer to the following files for more usages: - ``_ - ``_ tokensencoderdecoderjoiner keywords_file num_threads sample_rate feature_dimmax_active_pathskeywords_scorekeywords_thresholdnum_trailing_blanksproviderdevicec &t|t|t|t||dkDsJ|t|||}t| |}t||||}t ||}t ||| | | | |}t ||_y)a1 Please refer to ``_ to download pre-trained models for different languages, e.g., Chinese, English, etc. Args: tokens: Path to ``tokens.txt``. Each line in ``tokens.txt`` contains two columns:: symbol integer_id encoder: Path to ``encoder.onnx``. decoder: Path to ``decoder.onnx``. joiner: Path to ``joiner.onnx``. keywords_file: The file containing keywords, one word/phrase per line, and for each phrase the bpe/cjkchar/pinyin are separated by a space. num_threads: Number of threads for neural network computation. sample_rate: Sample rate of the training data used to train the model. feature_dim: Dimension of the feature used to train the model. max_active_paths: Use only when decoding_method is modified_beam_search. It specifies the maximum number of active paths during beam search. keywords_score: The boosting score of each token for keywords. The larger the easier to survive beam search. keywords_threshold: The trigger threshold (i.e. probability) of the keyword. The larger the harder to trigger. num_trailing_blanks: The number of trailing blanks a keyword should be followed. Setting to a larger value (e.g. 8) when your keywords has overlapping tokens between each other. provider: onnxruntime execution providers. Valid values are: cpu, cuda, coreml. device: onnxruntime cuda device index. r)rrr)r r!) transducerrrprovider_config) sampling_rater) feat_config model_configrrrrrN)rr r rrr_KeywordSpotterkeyword_spotter)selfrrrrrrrrrrrrr r!transducer_configr$r'r&keywords_spotter_configs r__init__zKeywordSpotter.__init__Qs~ F#G$G$F#Q+ +7  )  )(#+  -%# #7#%- 3)1'#  //FGrsc:|jj|y)zReset the stream after a keyword is detected. You should call this right after a keyword is detected and before feeding more audio to the stream. Args: s: The stream to be reset. N)r)resetr*r.s r reset_streamzKeywordSpotter.reset_streams ""1%rNkeywordscp||jjS|jj|S)arCreate a new stream for keyword spotting. Args: keywords: Optional extra keywords to add for this stream. The format is the same as the keywords file content. Use ``None`` to use only the keywords from the keywords file provided to the constructor. Returns: A new ``OnlineStream`` object. )r) create_stream)r*r3s rr5zKeywordSpotter.create_streams6  ''557 7''55h? ?rc:|jj|y)zlDecode one step for the given stream. Args: s: The stream to decode. N)r) decode_streamr1s rr7zKeywordSpotter.decode_streams **1-rssc:|jj|y)z{Decode on multiple streams at the same time. Args: ss: A list of streams to decode. N)r)decode_streams)r*r8s rr:zKeywordSpotter.decode_streamss ++B/rreturnc8|jj|S)zCheck whether the stream has enough frames for decoding. Args: s: The stream to check. Returns: ``True`` if the stream has enough frames for decoding. ``False`` otherwise. )r)is_readyr1s rr=zKeywordSpotter.is_readys##,,Q//rch|jj|jjS)aGet the keyword spotting result as a string. Args: s: The stream to get the result from. Returns: A string containing the detected keyword. Returns an empty string if no keyword is detected. )r) get_resultkeywordstripr1s rr?zKeywordSpotter.get_results*##..q199??AArcL|jj|jS)zGet the token list of the keyword result. Args: s: The stream to get the tokens from. Returns: A list of strings, each being a token. )r)r?rr1s rrzKeywordSpotter.tokenss!##..q1888rcL|jj|jS)zGet the timestamp list of the keyword result. Args: s: The stream to get the timestamps from. Returns: A list of floats, each being the timestamp (in seconds) of the corresponding token. )r)r? timestampsr1s rrDzKeywordSpotter.timestamps s!##..q1<<Pg?g?cpur)N)__name__ __module__ __qualname____doc__strintfloatr-r r2rr5r7rr:boolr=r?rrDrrr r se8B" ! #$(#$fHfHfH fH  fH  fHfHfHfHfHfH"fH!fHfHfHP &l &@hsm@ .|.0l!30 0, 04 0 BL BS B 9 9c 9 =L =T%[ =rr N)pathlibrtypingrrsherpa_onnx.lib._sherpa_onnxrrrr r r r r(rNrobjectrRrrrWs7!K434~=V~=r