Skip to content

(Code Mirror) A CLI tool to query and install different language voices for piper speech synthesizer (https://github.com/rhasspy/piper)

License

Notifications You must be signed in to change notification settings

think-biq/piper-whistle

Repository files navigation

piper-whistle

gitlab-pipieline-release-name gitlab-pipeline-release-status readthedocs-status

Tool to manage voices used with the piper speech synthesizer. You may also browse the docs online at https://think-biq.gitlab.io/piper-whistle/ There is also a quick guide on how to setup and use piper and (piper-)whistle.

usage

usage: piper_whistle [-h] [-d] [-v] [-V] [-P DATA_ROOT] [-R]
                     {refresh,guess,path,speak,list,preview,install,remove} ...

positional arguments:
  {refresh,guess,path,speak,list,preview,install,remove}

options:
  -h, --help            Show help message.
  -d, --debug           Activate very verbose logging.
  -v, --verbose         Activate verbose logging.
  -V, --version         Show version number.
  -P DATA_ROOT, --data-root DATA_ROOT
                        Root path where whistle should store config and data in.
  -R, --refresh         Refreshes (or sets up) language index by downloading the latest lookup.

commands

guess

usage: piper_whistle guess [-h] [-v] language_name

positional arguments:
  language_name  A string representing a language name (or code).

options:
  -h, --help     Show help message.
  -v, --verbose  Activate verbose logging.

Tries to guess the language you are looking for (and is supported by piper) from the name you provide.

path

usage: piper_whistle path [-h] [-v] voice_selector

positional arguments:
  voice_selector  Selector of voice to search.

options:
  -h, --help      show this help message and exit
  -v, --verbose   Activate verbose logging.

Shows the local path to a specific model. The voice_selector has the format:

${CODE}:${NAME}@${QUALITY}/${SPEAKER}

The ${SPEAKER} part is optional; as is the ${CODE} part. So if you want to select the voice named 'alba' in quality 'medium', you could simply query: alba@medium

The language code is infered. Alternatively, you can just query with the model name listed by the list command.

${CODE}-${NAME}-${QUALITY}

So for the example above, that would be en_GB-alba-medium

speak

usage: piper_whistle speak [-h] [-c CHANNEL] [-j] [-r] [-o OUTPUT] [-v] something

positional arguments:
  something             Something to speak.

options:
  -h, --help            Show help message.
  -c CHANNEL, --channel CHANNEL
                        Path to channel (named pipe (aka. fifo)) to which piper is listening.
  -j, --json            Encode the text as json payload. Is on by default.
  -r, --raw             Encode the text directly.
  -o OUTPUT, --output OUTPUT
                        Instead of streaming to audio channel, specifies a path to wav file where speech will be store in.
  -v, --verbose         Activate verbose logging.

Currently only works on linux / bsd systems, with a FIFO (aka. named pipes) setup. The basic idea is, having one pipe accepting json input (provided by this command), which is listened to by piper. After piper has processed the audio, it is either saved to file or passed on to another FIFO, which can then be read by a streaming audio player like aplay.

Example: Assuming piper is installed at /opt/wind/piper, the named pipes are located at /opt/wind/channels and whistle is available in $PATH, the aformentioned setup could look like the following:

pipes:

  • /opt/wind/channeld/speak - accepts json payload
  • /opt/wind/channeld/input - read by piper
  • /opt/wind/channeld/ouput - written by piper

processes:

  • tty0: tail -F /opt/wind/channels/speak | tee /opt/wind/channels/input
  • tty1: /opt/wind/piper/piper -m $(piper_whistle path alba@medium) --debug --json-input --output_raw < /opt/wind/channels/input > /opt/wind/channels/output
  • tty2: aplay --buffer-size=777 -r 22050 -f S16_LE -t raw < channels/output

The tail command makes sure, that the payload on speak is send to input, thereby keeping the file open after processing. Otherwise, the setup would exit after piper has finished the first payload. This way you can continually prompt.

list

usage: piper_whistle list [-h] [-v] [-I] [-a] [-L] [-g] [-U] [-S] [-p]
                          [-l LANGUAGE_CODE] [-i VOICE_INDEX]

options:
  -h, --help            Show help message.
  -v, --verbose         Activate verbose logging.
  -I, --installed       Only list installed voices.
  -a, --all             List voices for all available languages.
  -L, --languages       List available languages.
  -g, --legal           Show avaiable legal information.
  -U, --show-url        Show URL of voice on remote host.
  -S, --omit-speakers   Omit speakers form listing.
  -p, --install-path    Show path of voice (if installed).
  -l LANGUAGE_CODE, --language-code LANGUAGE_CODE
                        Only list voices matching this language.
  -i VOICE_INDEX, --voice-index VOICE_INDEX
                        List only specific language voice.

This command lets you investigate available voices for specific languages, or simply list all available voices. Using the --installed switch, you can filter voices that are currently installed in the local cache directory. The cache is located in the user app path, as provided by userpaths pip package. On linux this would be ${HOME}/.config/piper-whistle. You may also get the model path on the remote host using -U.

preview

usage: piper_whistle preview [-h] [-v] [-l LANGUAGE_CODE] [-i VOICE_INDEX]
                             [-s SPEAKER_INDEX] [-D]

options:
  -h, --help            Show help message.
  -v, --verbose         Activate verbose logging.
  -l LANGUAGE_CODE, --language-code LANGUAGE_CODE
                        Select language.
  -i VOICE_INDEX, --voice-index VOICE_INDEX
                        Specific language voice. (defaults to first one)
  -s SPEAKER_INDEX, --speaker-index SPEAKER_INDEX
                        Specific language voice speaker. (defaults to first one)
  -D, --dry-run         Build URL and simulate download.

With preview, you can download and play samples audio files, for any voice supported by piper. It currently uses mplayer to play the audio file.

install

usage: piper_whistle install [-h] [-v] [-D] language_code voice_index

positional arguments:
  language_code  Select language.
  voice_index    Specific language voice. (defaults to first one)

options:
  -h, --help     Show help message.
  -v, --verbose  Activate verbose logging.
  -D, --dry-run  Simulate download / install.

With install you can fetch available voice models and store them locally for use with piper. You may first want to search for a voice you like with list and then note the language code and index, so install knows where to look. The model file (onnx) as well as its accompanying config (json) file, will be stored in the local user data path as provide by userpaths. On linux this would be ${HOME}/.config/piper-whistle.

remove

usage: piper_whistle remove [-h] [-v] voice_selector

positional arguments:
  voice_selector  Selector of voice to search.

options:
  -h, --help      Show help message.
  -v, --verbose   Activate verbose logging.

Any installed voice model can be deleted, via remove. You may pass the model name or shorthand selector.