Skip to content

Latest commit

 

History

History
164 lines (111 loc) · 5.31 KB

README.md

File metadata and controls

164 lines (111 loc) · 5.31 KB
Raw

VoiceKit Python examples

You will need Python >= 3.5 to run this examples. To work correctly all scripts should be run from this directory (voicekit-examples/python).

Usage

Install the requirements

$ python3 -m pip install -r requirements.txt

You may install optional dependencies (opuslib and PyAudio) to gain additional functionality:

$ python3 -m pip install -r requirements/all.txt

Basic recognition examples

Run basic (non-streaming) speech recognition example:

$ python3 recognize.py -r 16000 -c 1 -e MPEG_AUDIO ../audio/sample_1.mp3

To disable automatic punctuation and get up to 3 recognition alternatives:

$ python3 recognize.py -r 16000 -c 1 -e MPEG_AUDIO --disable_automatic_punctuation --max_alternatives 3 ../audio/sample_1.mp3

Denormalization examples:

Enable:

$ python3 recognize.py -r 48000 -c 1 -e MPEG_AUDIO ../audio/sample_4.mp3

Disable:

$ python3 recognize.py -r 48000 -c 1 -e MPEG_AUDIO --disable_denormalization ../audio/sample_4.mp3

Profanity filter example: Enabled by default:

$ python3 recognize.py -r 16000 -c 1 -e MPEG_AUDIO ../audio/sample_5.mp3

Disable:

$ python3 recognize.py -r 16000 -c 1 -e MPEG_AUDIO --disable_profanity_filter ../audio/sample_5.mp3

Also, there is a REST-like API for non-streaming speech recognition (if you are planning to use REST in production, which is not recommended by the way, make sure to use libraries which support HTTP2, like httpx):

$ python3 recognize_rest.py -r 16000 -c 1 -e MPEG_AUDIO ../audio/sample_2.mp3

To get description of all command-line parameters:

$ python3 recognize.py --help

Advanced recognition examples

Use RAW_OPUS to save bandwidth:

$ python3 recognize_stream.py -e RAW_OPUS -r 48000 -c 1 output.raw_opus

Run streaming speech recognition with interim results and disabled voice activity detection (VAD):

$ python3 recognize_stream.py -e MPEG_AUDIO --interim_results --do_not_perform_vad -r 16000 -c 1 ../audio/sample_1.mp3

Use audio from microphone (requires PyAudio):

$ python3 recognize_stream.py -e LINEAR16 --interim_results -r 16000 -c 1 --silence_duration_threshold 0.3 pyaudio:

Specify longer silence timeout for voice activity detection:

$ python3 recognize_stream.py -e LINEAR16 --interim_results -r 16000 -c 1 --silence_duration_threshold 1.2 pyaudio:

Return just the first recognized utterance and halt:

$ python3 recognize_stream.py -e LINEAR16 --interim_results -r 16000 -c 1 --single_utterance pyaudio:

Speech synthesis examples

To run basic speech synthesis and save result to wav:

$ python3 synthesize_stream.py -r 48000 -e LINEAR16 "И мысли тоже тяжелые и медлительные, падают неторопливо и редко одна за другой, точно песчинки в разленившихся песочных часах." output_1.wav

You can also specify alternative stress with 0 sign after a vowel:

$ python3 synthesize_stream.py -r 48000 -e LINEAR16 "За0мок - замо0к." output_2.wav

Feel free to use arabic numerals and named entities:

$ python3 synthesize_stream.py -r 48000 -e LINEAR16 "Газета Times, 03 января 2009 года - Канцлер на грани ради второго спасения банков." output_3.wav

For now, LINEAR16 does not support samples rates other than 48kHz. Use RAW_OPUS (opuslib required) to specify different sample rates:

$ python3 synthesize_stream.py -r 16000 -e RAW_OPUS "Привет, мир." output_4.wav

It is possible to use SSML:

$ python3 synthesize_stream.py -r 48000 -e LINEAR16 --ssml "<speak><p><s>Оригинальная мысль?</s><s>Нет ничего легче.</s></p><break time='300ms'/><p><s>Библиотеки просто набиты ими.</s></p></speak>" output_5.wav

You can also specify voice name:

$ python3 synthesize_stream.py -r 48000 -e LINEAR16 --voice alyona "Привет! Я Алёна. Я помогу в озвучке книг, новостей, образовательных курсов, а также могу быть твоим напарником для медитации." output_6.wav

Synthesize and play audio directly through your speakers at the same time. This is faster than saving to file because of streaming (requires PyAudio):

$ python3 synthesize_stream.py -r 48000 -e LINEAR16 "И мысли тоже тяжелые и медлительные, падают неторопливо и редко одна за другой, точно песчинки в разленившихся песочных часах." pyaudio:

To get description of all command-line parameters:

$ python3 synthesize_stream.py --help

Generate Protobuf and gRPC definitions (optional)

In case of API changes (*.proto files in apis directory), you may regenerate Protobuf and gRPC definitions by simply running the following script (no need to re-clone the whole repo):

$ ./generate_protobuf.sh