Adds --server and --volume option

[flatsiedatsie]

This runs the core of the project in a loop until the process receives a termination signal.

After creating this issue I thought I'd see if I could add "server" functionality myself. It's probably not optimal, since I'm not a C++ coder. The bool running = true part might not be needed.

[flatsiedatsie]

The second commit to this pull request adds a -v or --volume option, which lets users set the desired output volume of the audio.

This also works with JSON input, to change the volume while in server mode. E.g.

echo '{"text":"Bla bla bla","volume_level":"0.1"}' | ./piper --model /home/pi/tts/en_US-lessac-medium.onnx --json-input --server --output-raw -volume 0.8 --sentence_silence 0.1 | aplay -D plughw:0,0 -r22050 -fS16_LE -t raw -

[akhil3417]

how would on go about using this server(i.e make requests)? how's this different from https://github.com/rhasspy/piper/blob/master/src/python_run/README_http.md

[flatsiedatsie]

It's not really a server, it's a long-running cli process.

• Useful when using cli is prefered over server route (e.g. avoid port use / manage all sub-processes similarly)
• Output can be piped directly to aplay as it is generated

Making a request works as follows:

if self.llm_tts_process.poll() == None:
    if self.DEBUG:
        print("llm_speak: LLM TTS Process is running")
    json_voice_message = '{ "text": "' + str(voice_message).replace('"', '\\"') + '","volume_level":"' + str( int(self.persistent_data['speaker_volume'])/100) + '"}\n'
    if self.DEBUG: 
        print("piping json into Piper: " + str(json_voice_message))
    self.llm_tts_process.stdin.write(json_voice_message)
    self.llm_tts_process.stdin.flush()

Looking back, --server is probably the wrong name for this feature if a real server already exists.

[akhil3417]

thanks, since the piper-tts python package doesn't sound as good (since the binary uses patched espeak-ng), this is alternative for now. can you share the binary ?

[flatsiedatsie]

Sure, the binary can be found here: https://github.com/createcandle/voco/tree/master/llm/tts

Implementation example can be found here:
https://github.com/createcandle/voco/blob/master/pkg/voco_adapter.py

[akhil3417]

linked binary was aarch64, forgot to mention but m running x86_64 arch.

exec: Failed to execute process: './piper' the file could not be run by the operating system.

[flatsiedatsie]

Then you'll have to compile it yourself.

[manoflinux]

this will increase the volume
echo "$texttoberead" | /home/user/tts/piper/piper/piper --model /home/user/tts/piper/voices/en_GB-cori-high.onnx --output-file -|ffmpeg -f wav -i - -af "volume=8dB" -f wav pipe:1 | aplay

[schmidt9]

@flatsiedatsie thank you for --server option, running Piper as long-running process using this option gives best performance compared to other variants (running as an ordinary command or using as http server), because it lets to load the model only once and reuse it and it has no overhead due to http server requests

[schmidt9]

The only downside now I cannot determine when the audio piped to aplay ends playing