tts-generator

name	tts-generator
description	Генерация голосовых промптов для MikoPBX модулей на 25+ языках через Piper TTS / MMS-TTS. Использовать при создании модулей с голосовыми сообщениями, добавлении новых языковых фраз или обновлении существующих звуковых файлов.
allowed-tools	Bash, Read, Write, Grep, Glob, Agent

MikoPBX TTS Voice Prompt Generator

You are a voice prompt generation assistant for MikoPBX. You generate WAV audio files from text in 25+ languages using TTS engines running on a remote Mac Studio.

When to Use

• User asks to generate audio/voice prompts for a module
• User needs custom phrases in multiple languages (e.g., "call direction prohibited")
• User wants to add sounds to a Language Pack module
• User mentions "сгенерируй аудио", "generate voice", "TTS", "голосовые подсказки"

Architecture

Local machine (development)
    ↓ SSH + rsync
Mac Studio VM: admin@thai-mac-studio-agent-01
    ~/Development/mikopbx-tts-generator/
    ├── generate_tts.py          # TTS generation script
    ├── {lang}/core-sounds-*.txt # Translation files
    └── output/{lang}/*.wav      # Generated audio

TTS Engines:

• Piper TTS (fast, ~0.6s/file, 22050Hz) — pt-pt, uk-ua, vi-vn, ro-ro, hu-hu, fi-fi, ka-ge, zh-cn, hr-hr(sr voice)
• MMS-TTS (Meta, 16000Hz, slower) — az-az, th-th
• Piper voices for other languages — ru-ru, en-us, en-gb, de-de, fr-fr, es-es, it-it, pl-pl, cs-cz, da-dk, nl-nl, sv-se, tr-tr, el-gr, pt-br

Supported Languages (25)

Built-in (human recordings exist, TTS for custom phrases only)

Code	Language	Piper Voice
en-en	English	en_GB-alba-medium
ru-ru	Russian	ru_RU-irina-medium

Language Pack modules (with native sounds)

Code	Language	Piper Voice
cs-cs	Czech	cs_CZ-jirka-medium
da-dk	Danish	da_DK-talesyntese-medium
de-de	German	de_DE-thorsten-medium
nl-nl	Dutch	nl_NL-mls-medium
es-es	Spanish	es_ES-sharvard-medium
fr-ca	French	fr_FR-siwis-medium
gr-gr	Greek	el_GR-rapunzel-low
it-it	Italian	it_IT-riccardo-x_low
pl-pl	Polish	pl_PL-darkman-medium
pt-br	Brazilian Portuguese	pt_BR-faber-medium
sv-sv	Swedish	sv_SE-nst-medium
tr-tr	Turkish	tr_TR-dfki-medium

Language Pack modules (TTS-generated)

Code	Language	Engine	Voice
pt-pt	Portuguese	Piper	pt_PT-tugão-medium
uk-ua	Ukrainian	Piper	uk_UA-ukrainian_tts-medium
vi-vn	Vietnamese	MMS	vie
ro-ro	Romanian	Piper	ro_RO-mihai-medium
hu-hu	Hungarian	Piper	hu_HU-anna-medium
fi-fi	Finnish	Piper	fi_FI-harri-medium
ka-ge	Georgian	Piper	ka_GE-natia-medium
zh-cn	Chinese	Piper	zh_CN-huayan-medium
az-az	Azerbaijani	MMS	azj-script_latin
th-th	Thai	MMS	tha
hr-hr	Croatian	Piper	sr_RS-serbski_institut-medium

Workflow for Custom Module Phrases

Step 1: Translate the phrase

For each target language, translate the phrase. Use Claude to translate directly:

English: "This call direction is prohibited"
Russian: "Данное направление звонка запрещено"
Ukrainian: "Цей напрямок дзвінка заборонено"
Georgian: "ზარის ეს მიმართულება აკრძალულია"
... (all 25 languages)

Step 2: Create translation file on VM

Write a temporary file on the VM:

ssh admin@thai-mac-studio-agent-01 'cat > ~/Development/mikopbx-tts-generator/custom-phrase.txt << EOF
; Custom phrase for ModuleUsersGroups
call-direction-prohibited: <translated text>
EOF'

Step 3: Generate audio

For Piper languages:

ssh admin@thai-mac-studio-agent-01 'eval "$(/opt/homebrew/bin/brew shellenv)" && \
  echo "<translated text>" | piper \
    --model ~/.local/share/piper-voices/<voice>.onnx \
    --output_file ~/Development/mikopbx-tts-generator/output/<lang>/call-direction-prohibited.wav'

For MMS languages:

ssh admin@thai-mac-studio-agent-01 'eval "$(/opt/homebrew/bin/brew shellenv)" && \
  cd ~/Development/mikopbx-tts-generator && \
  python3.11 generate_tts.py --lang <lang> --key call-direction-prohibited --format native'

Step 4: Copy to module

rsync -avz admin@thai-mac-studio-agent-01:~/Development/mikopbx-tts-generator/output/<lang>/call-direction-prohibited.wav \
  /path/to/Module/Sounds/<lang>/call-direction-prohibited.wav

Batch Generation (generate_tts.py)

The main script handles batch generation from translation files:

# Preview what would be generated
ssh admin@thai-mac-studio-agent-01 'eval "$(/opt/homebrew/bin/brew shellenv)" && \
  cd ~/Development/mikopbx-tts-generator && \
  python3.11 generate_tts.py --lang uk-ua --dry-run'

# Generate single file
python3.11 generate_tts.py --lang uk-ua --key vm-goodbye --format native

# Generate all files for a language
python3.11 generate_tts.py --lang uk-ua --format native

# Generate all languages
python3.11 generate_tts.py --lang all --format native

Format options

• native — TTS native rate (22kHz Piper / 16kHz MMS) — recommended
• 8k — 8000 Hz mono (classic Asterisk)
• 16k — 16000 Hz mono (wideband)
• 22k — 22050 Hz mono

Quick Single Phrase Generation

For generating a single phrase in one language without the full script:

# Piper (most languages)
ssh admin@thai-mac-studio-agent-01 'eval "$(/opt/homebrew/bin/brew shellenv)" && \
  echo "Данное направление звонка запрещено" | piper \
    --model ~/.local/share/piper-voices/ru_RU-irina-medium.onnx \
    --output_file /tmp/test-phrase.wav'

# Copy result
scp admin@thai-mac-studio-agent-01:/tmp/test-phrase.wav /tmp/

Voice Model Management

Models are auto-downloaded on first use to ~/.local/share/piper-voices/. To pre-download:

ssh admin@thai-mac-studio-agent-01 'eval "$(/opt/homebrew/bin/brew shellenv)" && \
  cd ~/.local/share/piper-voices && \
  curl -sL "https://huggingface.co/rhasspy/piper-voices/resolve/main/<lang_short>/<lang_code>/<voice_name>/<quality>/<voice_id>.onnx" -o "<voice_id>.onnx" && \
  curl -sL "https://huggingface.co/rhasspy/piper-voices/resolve/main/<lang_short>/<lang_code>/<voice_name>/<quality>/<voice_id>.onnx.json" -o "<voice_id>.onnx.json"'

Output Format Notes

• WAV files are 22050Hz (Piper) or 16000Hz (MMS), mono, 16-bit PCM
• MikoPBX auto-converts to all Asterisk formats on module install (ulaw, alaw, gsm, g722, sln)
• Higher source quality = better output on HD codecs (Opus, G.722)
• No need to convert manually — just provide WAV

Module Sound File Structure

ModuleName/
  Sounds/
    <asterisk-lang-code>/
      custom-phrase.wav
      another-phrase.wav
      subfolder/
        nested-phrase.wav

Language Pack modules use the Asterisk language code as directory name (e.g., uk-ua, pt-pt). Feature modules can use any structure — files get prefixed with module name on install.

Licensing

• Module code: GPL-3.0-or-later (MikoPBX standard)
• Sound files: CC BY-SA 4.0 (compatible with Asterisk sounds)
• TTS engines: Piper (MIT), MMS-TTS (CC BY-NC 4.0 model — output not restricted)
• Language packs are free for users to install and download