Qwen3 Tts Hindi LoRA Grpo

--- license: apache-2.0 language: - hi library_name: mlx tags: - mlx - lora - grpo - rl - tts - hindi - text-to-speech - apple-silicon base_model: mlx-community/Qwen3-TTS-12Hz-0.6B-Base-8bit pipeline_tag: text-to-speech --- # Qwen3-TTS Hindi LoRA — GRPO-refined (Pipeline 3) A Hindi LoRA adapter for [Qwen3-TTS-0.6B](https://huggingface.co/mlx-community/Qwen3-TTS-12Hz-0.6B-Base-8bit), **post-trained with GRPO** (reinforcement learning) on top of the supervised [akashicmarga/qwen3-tts-hindi-lora](https://huggingface.co/akashicmarga/qwen3-tts-hindi-lora) (SFT) adapter. SFT optimizes next-token cross-entropy, which never directly targets *intelligibility*. GRPO optimizes it directly: the model samples speech, an ASR (whisper-large-v3-turbo) transcribes it, and **CER vs the target text** becomes the reward — so the policy is nudged toward audio a listener (proxied by ASR) can actually understand, while a KL penalty keeps it close to the SFT model. ## Method | | | |---|---| | Algorithm | GRPO (group-relative, critic-free), strictly on-policy | | Reward | intelligibility `1 − min(1, CER)` (whisper-large-v3-turbo) + length/degeneracy guard | | Rollout layout | **interleaved** — matches the real `generate()` path (essential; the SFT-concatenated variant did not transfer) | | KL reference | frozen SFT adapter snapshot, β = 0.08 (k3 estimator) | | Prompts | 863 Hindi "Read"-scenario sentences (IndicVoices-R) | | Steps | 400 (plateaued ~step 200), LR 5e-6, group size 4 | | Trainable | rank-8 LoRA on talker q/k/v/o/gate/up/down (462 tensors) | ## Evaluation (120 held-out sentences, 2 seeds, CER capped at 1.0) | | CER ↓ | DNSMOS OVRL ↑ | SIG ↑ | |---|---|---|---| | SFT baseline | 0.205 | 3.237 | 3.573 | | **GRPO (this model)** | **0.183** | **3.271** | **3.608** | - **CER −11% relative**, better on 75/120 sentences (Wilcoxon p ≈ 7e-4; paired t p ≈ 0.09). - **Naturalness (DNSMOS) small but significant gain** (OVRL/SIG, p < 0.05) — no quality trade-off for the intelligibility gain. A modest but statistically robust improvement. (DNSMOS is English-trained, so a rough relative proxy for Hindi naturalness; CER is an ASR proxy, not human MOS. GRPO reduces typical-case errors but does not eliminate occasional non-terminating hallucinations.) ## Samples — SFT vs GRPO Same sentence, same seed — only the adapter differs. These are normal-length held-out commands (no le...