darijavan/malagasy-voice

Malagasy TTS & Voice AI — text-to-speech + speech recognition pipeline (fine-tuned, low-resource). Beginner-to-production learning project. # Malagasy TTS & Voice AI Text-to-speech **and** speech recognition for Malagasy (low-resource language), built by fine-tuning pretrained models on a self-collected dataset. > Beginner-to-production learning project. Stack: Python, PyTorch, Coqui TTS, > Whisper, audio DSP. Compute: 1× RTX 2070 (8 GB) via WSL2. ## Scope (decided 2026-07-12) - **Both TTS + ASR** (matches the year goal). - **Standalone first**; Vavaka Flutter app integration is deferred to a later phase. - Dataset: to be recorded (Malagasy read speech, ~1–2h baseline). ## Reality check The model is the easy 20%; **the dataset is the hard 80%**. Malagasy has no public TTS/ASR corpus, so the real work is data engineering + fine-tuning, not training from scratch. Fine-tuning is the correct choice on a single 8 GB GPU. ## Repo layout ``` data/raw/ # original recordings (gitignored) data/processed/ # aligned, cleaned, phonemized pairs (gitignored) notebooks/ # exploration + DSP learning src/ # reusable pipeline code scripts/ # CLI entry points checkpoints/ # model weights (gitignored) outputs/ # generated audio / transcripts (gitignored) ``` ## Learning path (sequenced for a beginner) 0. Env: Python venv/uv, PyTorch+CUDA, librosa, Coqui TTS, Whisper 1. DL fundamentals (tensors, autograd, training loops) — see courses-site AI curriculum 2. Audio DSP (STFT, mel-spectrograms, normalization) 3. TTS architectures (VITS, XTTS/YourTTS, Piper) 4. ASR (Whisper fine-tuning, CTC, decoding) 5. Data pipeline ⭐ (forced alignment, Malagasy G2P, cleaning) 6. Low-resource tricks (augmentation, transfer learning, few-shot cloning) ## Quick start ```bash python3 -m venv .venv && source .venv/bin/activate pip install -r requirements.txt python -c "import torch; print(torch.cuda.is_available())" # should print True ``` See the Notion "Malagasy TTS & Voice AI" project for the full task breakdown.