Morisyen TTS
Text to speech for Kreol Morisien
Generate Mauritian Creole speech for applications, conversational interfaces, and audio content.
Overview
Morisyen TTS converts written Kreol Morisien (Mauritian Creole) into speech. It supports complete audio responses through HTTP and incremental audio delivery through WebSocket.
| Specification | Value |
|---|---|
| Language | Kreol Morisien |
| Language code | mfe |
| Input | UTF-8 text |
| Native audio sample rate | 24,000 Hz |
| Documented API audio | Mono, 8,000 Hz, 16-bit PCM WAV |
| Interfaces | HTTP and WebSocket |
| Checkpoint format | Safetensors |
| License | Proprietary — non-commercial use only |
Quick start
The examples connect to a running Morisyen TTS service. Set MORISYEN_TTS_URL to the service endpoint available to you; this repository does not provide a hosted endpoint.
export MORISYEN_TTS_URL="http://localhost:8001"
curl --fail-with-body "$MORISYEN_TTS_URL/synthesize" \
-H "Content-Type: application/json" \
-d '{"gen_text":"Bonzur! Kouma ou ete zordi?"}'
The response contains a base64-encoded WAV file. With a copy of this repository, save it directly using the included Python client:
python examples/synthesize.py "Bonzur! Kouma ou ete zordi?" --output bonzur.wav
For Python applications:
from client import MorisyenTTSClient
client = MorisyenTTSClient(base_url="http://localhost:8001")
with open("bonzur.wav", "wb") as audio:
audio.write(client.synthesize_wav("Bonzur! Kouma ou ete zordi?"))
See Usage for streaming and checkpoint downloads.
Audio interfaces
- HTTP: receive a complete WAV file encoded as base64 in a JSON response.
- WebSocket: receive successive WAV chunks while the request is processed.
- Python client: choose full WAV files or raw PCM16 bytes for playback integration.
The documented API returns 8 kHz audio. Native model audio and API output have different sample rates; use the rate of the audio you actually receive.
Repository structure
morisyen-tts/
├── README.md
├── LICENSE
├── client/
│ ├── __init__.py
│ └── morisyen_tts.py
├── docs/
│ ├── API.md
│ ├── LIMITATIONS.md
│ └── USAGE.md
├── examples/
│ ├── stream.py
│ └── synthesize.py
├── scripts/
│ └── MorisyenTTSWS.py
├── requirements-client.txt
└── weights/
└── t3_finetuned_merged.safetensors
client/ is the application integration entry point. examples/ contains runnable clients. scripts/MorisyenTTSWS.py preserves the previous client import location.
Checkpoint access
The checkpoint is available in weights/. It requires a compatible inference runtime; the Python clients in this repository communicate with an existing service and do not load model weights. A download command is provided in Usage.
Intended use and limitations
Permitted uses include non-commercial Kreol Morisien audio content, educational narration, and research or evaluation of conversational interfaces, subject to the license. Review pronunciation for names, numbers, abbreviations, and mixed-language text. For longer passages, use sentence-sized requests and review the complete output.
Speech quality varies with the text and voice configuration. No independently reproducible benchmark or guaranteed latency is published with this release. See Limitations.
Feedback
Report pronunciation and integration issues through the repository's Discussions. Include a short, non-sensitive example and the behavior you expected.
License
Morisyen TTS is distributed under a proprietary, non-commercial license. Only personal, educational, and non-commercial research or evaluation use is permitted. Commercial use is prohibited, including use in business operations, paid services, and monetized products or content. Redistribution requires prior written permission. Third-party materials retain their respective licenses.