larynx 0.3.1

End-to-end text to speech using IPA and onnx models

Larynx

End-to-end text to speech system using gruut and onnx.

Larynx's goals are:

• "Good enough" synthesis to avoid using a cloud service
• Faster than realtime performance on a Raspberry Pi 4
• Broad language support
• Voices trained purely from public datasets

Samples

Listen to voice samples from all of the pre-trained models.

Installation

$ pip install larynx

For Raspberry Pi (ARM), you will first need to manually install phonetisaurus.

Language Download

Larynx uses gruut to transform text into phonemes. You must install the appropriate gruut language before using Larynx. U.S. English is included with gruut, but for other languages:

$ python3 -m gruut <LANGUAGE> download

Voice/Vocoder Download

Voices and vocoders are available to download from the release page. They can be extracted anywhere, and the directory simply needs to be referenced in the command-line (e,g, --glow-tts /path/to/voice).

Web Server

You can run a local web server with:

$ python3 -m larynx.server --voices-dir /path/to/voices

Visit http://localhost:5002 to view the site and try out voices. See http://localhost/5002/openapi for documentation on the available HTTP endpoints.

See --help for more options.

Command-Line Example

The command below synthesizes multiple sentences and saves them to a directory. The --csv command-line flag indicates that each sentence is of the form id|text where id will be the name of the WAV file.

$ cat << EOF |
s01|The birch canoe slid on the smooth planks.
s02|Glue the sheet to the dark blue background.
s03|It's easy to tell the depth of a well.
s04|These days a chicken leg is a rare dish.
s05|Rice is often served in round bowls.
s06|The juice of lemons makes fine punch.
s07|The box was thrown beside the parked truck.
s08|The hogs were fed chopped corn and garbage.
s09|Four hours of steady work faced us.
s10|Large size in stockings is hard to sell.
EOF
  larynx \
    --debug \
    --csv \
    --glow-tts local/en-us/harvard-glow_tts \
    --hifi-gan local/hifi_gan/universal_large \
    --output-dir wavs \
    --language en-us \
    --denoiser-strength 0.001

You can use the --interactive flag instead of --output-dir to type sentences and have the audio played immediately using sox.

GlowTTS Settings

The GlowTTS voices support two additional parameters:

• --noise-scale - determines the speaker volatility during synthesis (0-1, default is 0.333)
• --length-scale - makes the voice speaker slower (< 1) or faster (> 1)

Vocoder Settings

• --denoiser-strength - runs the denoiser if > 0; a small value like 0.005 is recommended.

Text to Speech Models

• GlowTTS (35 voices)
  • English (en-us, 20 voices)
    • blizzard_fls (F, accent, Blizzard)
    • cmu_aew (M, Arctic)
    • cmu_ahw (M, Arctic)
    • cmu_aup (M, accent, Arctic)
    • cmu_bdl (M, Arctic)
    • cmu_clb (F, Arctic)
    • cmu_eey (F, Arctic)
    • cmu_fem (M, Arctic)
    • cmu_jmk (M, Arctic)
    • cmu_ksp (M, accent, Arctic)
    • cmu_ljm (F, Arctic)
    • cmu_lnh (F, Arctic)
    • cmu_rms (M, Arctic)
    • cmu_rxr (M, Arctic)
    • cmu_slp (F, accent, Arctic)
    • cmu_slt (F, Arctic)
    • ek (F, accent, M-AILabs)
    • harvard (F, accent, CC/Attr/NC)
    • kathleen (F, CC0)
    • ljspeech (F, Public Domain)
  • German (de-de, 1 voice)
    • thorsten (M, CC0)
  • French (fr-fr, 3 voices)
    • gilles_le_blanc (M, M-AILabs)
    • siwis (F, CC/Attr)
    • tom (M, ODbL)
  • Spanish (es-es, 2 voices)
    • carlfm (M, public domain)
    • karen_savage (F, M-AILabs)
  • Dutch (nl, 3 voices)
    • bart_de_leeuw (M, Apache2)
    • flemishguy (M, CC0)
    • rdh (M, CC0)
  • Italian (it-it, 2 voices)
    • lisa (F, M-AILabs)
    • riccardo_fasol (M, Apache2)
  • Swedish (sv-se, 1 voice)
    • talesyntese (M, CC0)
  • Russian (ru-ru, 3 voices)
    • hajdurova (F, M-AILabs)
    • nikolaev (M, M-AILabs)
    • minaev (M, M-AILabs)
• Tacotron2
  • Coming soon

Vocoders

• Hi-Fi GAN
  • Universal large
  • VCTK medium
  • VCTK small
• WaveGlow
  • 256 channel trained on LJ Speech