Simple REST-style HTTP TTS server using

• MaryTTS, espeak and sequitur for g2p,
• MaryTTS and espeak for synthesis and
• pulseaudio for actual audio output.

Just run

./tts-server -h

to get help. The TTS server by default listens on localhost port 8300 if you specify no arguments.

• GET tts/g2p
• args:
  • l : 'en_US', 'de_DE', 'fr_FR', ...
  • v : voice to use (see below for examples)
  • e : 'mary', 'espeak' or 'sequitur'
  • t : text or ipa to synthesize, utf-8 encoded

Returns:

• 400 if request is invalid
• 200 OK {"ipa": "'ha\u028apt-ba\u02d0n-ho\u02d0f"}

Example:

curl -i 'http://localhost:8300/tts/g2p?l=de&v=de&e=sequitur&t=hauptbahnhof'

• GET tts/synth
• args:
  • l : 'en_US', 'de_DE', 'fr_FR', ...
  • v : voice to use (see below for examples)
  • e : 'mary' or 'espeak'
  • m : 'txt' or 'ipa'
  • t : text or ipa to synthesize, utf-8 encoded

Returns:

• 400 if request is invalid
• 200 OK, wave data

Example:

curl 'http://localhost:8300/tts/synth?l=de_DE&v=bits3&e=mary&m=txt&t=hauptbahnhof' >foo.wav

• POST tts/play
• args:
  • async : t to return immediately, otherwise: wait for playback to finish

Returns:

• 400 if request is invalid
• 200 OK otherwise

Example:

curl -H "Content-Type: audio/wav" -X POST --data-binary @foo.wav 'http://localhost:8300/tts/play?async=t'

Note: very incomplete.

• Python 2.7
• Mary TTS
• espeak
• sequitur g2p

Which voices are available in MaryTTS depends on your installation. Here are some typical examples:

• english, male

  • dfki-spike
  • dfki-obadiah
  • dfki-obadiah-hsmm
  • cmu-bdl-hsmm
  • cmu-rms-hsmm

• english, female

  • dfki-poppy
  • dfki-poppy-hsmm
  • dfki-prudence
  • dfki-prudence-hsmm
  • cmu-slt-hsmm

• german, male

  • dfki-pavoque-neutral
  • dfki-pavoque-neutral-hsmm
  • dfki-pavoque-styles
  • bits3
  • bits3-hsmm

• german, female

  • bits1-hsmm

• french, male

  • upmc-pierre-hsmm
  • upmc-pierre
  • enst-dennys-hsmm

• french, female

  • enst-camille
  • enst-camille-hsmm
  • upmc-jessica
  • upmc-jessica-hsmm

My own scripts as well as the data I create (i.e. lexicon and transcripts) is LGPLv3 licensed unless otherwise noted in the script's copyright headers.

Some scripts and files are based on works of others, in those cases it is my intention to keep the original license intact. Please make sure to check the copyright headers inside for more information.

Guenter Bartsch [email protected]