Update README.md
Browse files
README.md
CHANGED
|
@@ -1,22 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
- text-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
|
| 15 |
|
| 16 |
-
|
| 17 |
-
- **License:** apache-2.0
|
| 18 |
-
- **Finetuned from model :** Vyvo/VyvoTTS-LFM2-Neuvillette
|
| 19 |
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
-
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model:
|
| 3 |
+
- LiquidAI/LFM2-350M
|
| 4 |
+
tags:
|
| 5 |
+
- text-generation-inference
|
| 6 |
+
- transformers
|
| 7 |
+
- unsloth
|
| 8 |
+
- lfm2
|
| 9 |
+
- trl
|
| 10 |
+
license: apache-2.0
|
| 11 |
+
language:
|
| 12 |
+
- en
|
| 13 |
+
- ml
|
| 14 |
+
pipeline_tag: text-to-speech
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# Malayalam TTS Model (LFM2-350M Fine-tuned)
|
| 18 |
+
|
| 19 |
+
This repository contains a fine-tuned **Malayalam Text-to-Speech (TTS)** model based on **LFM2-350M**, trained using [VyvoTTS](https://github.com/Vyvo-Labs/VyvoTTS) (LLM-based TTS framework) and [Unsloth](https://github.com/unslothai/unsloth).
|
| 20 |
+
|
| 21 |
---
|
| 22 |
+
Malayalam TTS — 24 kHz (LLM + SNAC Codec)
|
| 23 |
+
|
| 24 |
+
High-quality Malayalam text-to-speech model targeting natural pronunciation and clean prosody at 24 kHz, using a discrete audio codec (SNAC 24 kHz) for waveform reconstruction. Designed for lightweight deployment (~350M parameters) with GPU/CPU support.
|
| 25 |
+
|
| 26 |
+
Status: v0.1 — stable inference, strong pronunciation, limited emotional expressiveness. Roadmap includes expressive styles and non‑verbal cues (laughter, giggles, breaths).
|
| 27 |
+
|
| 28 |
+
✨ Highlights
|
| 29 |
+
|
| 30 |
+
Language: Malayalam (with support for basic English loanwords).
|
| 31 |
+
|
| 32 |
+
Sample Rate: 24 kHz, mono.
|
| 33 |
+
|
| 34 |
+
Codec: [SNAC 24 kHz] for fast decoding.
|
| 35 |
+
|
| 36 |
+
Model Size: ~350M parameters (small/efficient).
|
| 37 |
|
| 38 |
+
Strengths: Clear, non‑robotic pronunciation; punctuation‑aware phrasing.
|
| 39 |
|
| 40 |
+
Known Limits: Emotion range is narrow; limited style transfer; no speaker cloning in v0.1.
|
|
|
|
|
|
|
| 41 |
|
| 42 |
+
## 📖 Model Details
|
| 43 |
+
- **Base Model:** LFM2-350M
|
| 44 |
+
- **Language:** Malayalam
|
| 45 |
+
- **Dataset:** [ai4bharat/rasa](https://huggingface.co/datasets/ai4bharat/rasa) (Malayalam subset)
|
| 46 |
+
- **Training:** 10 epochs, ~77k steps
|
| 47 |
+
- **Frameworks Used:** VyvoTTS, Unsloth
|
| 48 |
|
| 49 |
+
---
|