Upload 21 files

README.md

@@ -0,0 +1,92 @@
+---
+language:
+- saz
+- ta
+license: cc-by-4.0
+tags:
+- text-to-speech
+- tts
+- vits
+- sourashtra
+- low-resource
+pipeline_tag: text-to-speech
+---
+
+# Sourashtra VITS TTS Models
+
+VITS text-to-speech models for the [Sourashtra language](https://en.wikipedia.org/wiki/Sourashtra_language) (ISO 639-3: `saz`), a minority Indo-Aryan language spoken primarily in Tamil Nadu, India. Trained using [Coqui TTS](https://github.com/coqui-ai/TTS) on a custom annotated speech corpus.
+
+Four variants: **2 speakers** (male, female) × **2 input scripts** (Tamil script, Sourashtra script).
+
+---
+
+## Models
+
+| Folder | Speaker | Input Script | Training Steps |
+|--------|---------|--------------|----------------|
+| `Sourashtra-Male_Script-tamil` | Male | Tamil (தமிழ்) | 300,000 |
+| `Sourashtra-Male_Script-sourashtra` | Male | Sourashtra (ꢪꢾꢥꢶꢒ) | 300,000 |
+| `Sourashtra-Female_Script-tamil` | Female | Tamil (தமிழ்) | 340,000 |
+| `Sourashtra-Female_Script-sourashtra` | Female | Sourashtra (ꢪꢾꢥꢶꢒ) | 340,000 |
+
+Each folder contains `best_model.pth`, `config.json`, `inference.py`, and `requirements.txt`.
+
+---
+
+## Setup
+
+```bash
+pip install -r requirements.txt
+```
+
+For GPU inference, install the CUDA-enabled PyTorch build matching your driver first — see [pytorch.org](https://pytorch.org/get-started/locally/).
+
+---
+
+## Usage
+
+Run `inference.py` from inside the model folder:
+
+```bash
+# Male — Tamil script
+cd Sourashtra-Male_Script-tamil
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav
+
+# Male — Sourashtra script
+cd Sourashtra-Male_Script-sourashtra
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav
+
+# Female — Tamil script
+cd Sourashtra-Female_Script-tamil
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav
+
+# Female — Sourashtra script
+cd Sourashtra-Female_Script-sourashtra
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav
+```
+
+Use `--gpu <id>` to select a GPU, or `--cpu` to force CPU inference.
+
+---
+
+## Script Notes
+
+The Tamil-script and Sourashtra-script models produce speech from the same speaker — only the input orthography differs. Choose based on your text source.
+
+- **Tamil script models** — strip `:`, `.`, `'` and apply NFC normalization automatically
+- **Sourashtra script models** — strip Sourashtra Danda (꣎) and Double Danda (꣏) automatically
+
+---
+
+## Training
+
+| Parameter | Value |
+|-----------|-------|
+| Architecture | VITS (end-to-end, flow-based) |
+| Sample rate | 22050 Hz |
+| Mel bins | 80 |
+| Batch size | 16 |
+| Mixed precision | Yes |
+| Phonemes | No (character-level) |
+
+Male training data: ~9,800–10,000 utterances. Female training data: ~11,400 utterances.

Sourashtra-Female_Script-sourashtra/README.md

@@ -0,0 +1,44 @@
+---
+language:
+- saz
+license: cc-by-4.0
+tags:
+- text-to-speech
+- tts
+- vits
+- sourashtra
+- low-resource
+pipeline_tag: text-to-speech
+---
+
+# VITS TTS — Sourashtra Female (Sourashtra Script)
+
+VITS text-to-speech model for the [Sourashtra language](https://en.wikipedia.org/wiki/Sourashtra_language), a minority language spoken primarily in Tamil Nadu, India. This model uses **Sourashtra script** (ꢪꢾꢥꢶꢒ) input and synthesizes speech from a **female speaker**.
+
+## Model Details
+
+| Property | Value |
+|----------|-------|
+| Architecture | [VITS](https://arxiv.org/abs/2106.06103) (Coqui TTS) |
+| Speaker | Female |
+| Input script | Sourashtra (Unicode block U+A880–U+A8D9) |
+| Sample rate | 22050 Hz |
+| Training steps | 340,000 |
+
+## Setup
+
+```bash
+pip install -r requirements.txt
+```
+
+For GPU inference, install the CUDA-enabled PyTorch build matching your driver first — see [pytorch.org](https://pytorch.org/get-started/locally/).
+
+## Usage
+
+```bash
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav --gpu 1
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav --cpu
+```
+
+Input must be in Sourashtra script. For Tamil script input, use the `Sourashtra-Female_Script-tamil` model instead.

Sourashtra-Female_Script-sourashtra/best_model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bff515fc60c05aabcac5c7f4d7a22be2dc77e29521d9d8a7ee7790a7e715427
+size 997843894

Sourashtra-Female_Script-sourashtra/config.json

@@ -0,0 +1,261 @@
+{
+    "output_path": "./output_Sourashtra_Female_sourashtra",
+    "logger_uri": null,
+    "run_name": "vits_Sourashtra_Female_sourashtra",
+    "project_name": null,
+    "run_description": "\ud83d\udc38Coqui trainer run.",
+    "print_step": 25,
+    "plot_step": 100,
+    "model_param_stats": false,
+    "wandb_entity": null,
+    "dashboard_logger": "tensorboard",
+    "save_on_interrupt": true,
+    "log_model_step": null,
+    "save_step": 10000,
+    "save_n_checkpoints": 7,
+    "save_checkpoints": true,
+    "save_all_best": false,
+    "save_best_after": 0,
+    "target_loss": null,
+    "print_eval": true,
+    "test_delay_epochs": -1,
+    "run_eval": true,
+    "run_eval_steps": null,
+    "distributed_backend": "nccl",
+    "distributed_url": "tcp://localhost:54321",
+    "mixed_precision": true,
+    "precision": "fp16",
+    "epochs": 500,
+    "batch_size": 16,
+    "eval_batch_size": 8,
+    "grad_clip": [
+        1000,
+        1000
+    ],
+    "scheduler_after_epoch": true,
+    "lr": 0.001,
+    "optimizer": "AdamW",
+    "optimizer_params": {
+        "betas": [
+            0.8,
+            0.99
+        ],
+        "eps": 1e-09,
+        "weight_decay": 0.01
+    },
+    "lr_scheduler": null,
+    "lr_scheduler_params": null,
+    "use_grad_scaler": false,
+    "allow_tf32": false,
+    "cudnn_enable": true,
+    "cudnn_deterministic": false,
+    "cudnn_benchmark": false,
+    "training_seed": 54321,
+    "model": "vits",
+    "num_loader_workers": 8,
+    "num_eval_loader_workers": 4,
+    "use_noise_augment": false,
+    "audio": {
+        "fft_size": 1024,
+        "sample_rate": 22050,
+        "win_length": 1024,
+        "hop_length": 256,
+        "num_mels": 80,
+        "mel_fmin": 0,
+        "mel_fmax": null
+    },
+    "model_args": {
+        "num_chars": 97,
+        "out_channels": 513,
+        "spec_segment_size": 32,
+        "hidden_channels": 192,
+        "hidden_channels_ffn_text_encoder": 768,
+        "num_heads_text_encoder": 2,
+        "num_layers_text_encoder": 6,
+        "kernel_size_text_encoder": 3,
+        "dropout_p_text_encoder": 0.1,
+        "dropout_p_duration_predictor": 0.5,
+        "kernel_size_posterior_encoder": 5,
+        "dilation_rate_posterior_encoder": 1,
+        "num_layers_posterior_encoder": 16,
+        "kernel_size_flow": 5,
+        "dilation_rate_flow": 1,
+        "num_layers_flow": 4,
+        "resblock_type_decoder": "1",
+        "resblock_kernel_sizes_decoder": [
+            3,
+            7,
+            11
+        ],
+        "resblock_dilation_sizes_decoder": [
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ]
+        ],
+        "upsample_rates_decoder": [
+            8,
+            8,
+            2,
+            2
+        ],
+        "upsample_initial_channel_decoder": 512,
+        "upsample_kernel_sizes_decoder": [
+            16,
+            16,
+            4,
+            4
+        ],
+        "periods_multi_period_discriminator": [
+            2,
+            3,
+            5,
+            7,
+            11
+        ],
+        "use_sdp": true,
+        "noise_scale": 1.0,
+        "inference_noise_scale": 0.667,
+        "length_scale": 1,
+        "noise_scale_dp": 1.0,
+        "inference_noise_scale_dp": 1.0,
+        "max_inference_len": null,
+        "init_discriminator": true,
+        "use_spectral_norm_disriminator": false,
+        "use_speaker_embedding": false,
+        "num_speakers": 0,
+        "speakers_file": null,
+        "d_vector_file": null,
+        "speaker_embedding_channels": 256,
+        "use_d_vector_file": false,
+        "d_vector_dim": 0,
+        "detach_dp_input": true,
+        "use_language_embedding": false,
+        "embedded_language_dim": 4,
+        "num_languages": 0,
+        "language_ids_file": null,
+        "use_speaker_encoder_as_loss": false,
+        "speaker_encoder_config_path": "",
+        "speaker_encoder_model_path": "",
+        "condition_dp_on_speaker": true,
+        "freeze_encoder": false,
+        "freeze_DP": false,
+        "freeze_PE": false,
+        "freeze_flow_decoder": false,
+        "freeze_waveform_decoder": false,
+        "encoder_sample_rate": null,
+        "interpolate_z": true,
+        "reinit_DP": false,
+        "reinit_text_encoder": false
+    },
+    "_supports_cloning": false,
+    "use_phonemes": false,
+    "phonemizer": null,
+    "phoneme_language": "en-us",
+    "compute_input_seq_cache": true,
+    "text_cleaner": "multilingual_cleaners",
+    "enable_eos_bos_chars": false,
+    "test_sentences_file": "",
+    "phoneme_cache_path": "./output_Sourashtra_Female_sourashtra/phoneme_cache",
+    "characters": {
+        "characters_class": "TTS.tts.models.vits.VitsCharacters",
+        "vocab_dict": null,
+        "pad": "<PAD>",
+        "eos": "<EOS>",
+        "bos": "<BOS>",
+        "blank": "<BLNK>",
+        "characters": " !'(),-./12:?|\u00b3\u200d\ua882\ua883\ua884\ua885\ua886\ua887\ua888\ua889\ua88c\ua88d\ua88e\ua88f\ua890\ua891\ua892\ua893\ua894\ua895\ua896\ua897\ua898\ua899\ua89a\ua89b\ua89c\ua89d\ua89e\ua89f\ua8a0\ua8a1\ua8a2\ua8a3\ua8a4\ua8a5\ua8a6\ua8a7\ua8a8\ua8a9\ua8aa\ua8ab\ua8ac\ua8ad\ua8ae\ua8af\ua8b0\ua8b1\ua8b2\ua8b3\ua8b4\ua8b5\ua8b6\ua8b7\ua8b8\ua8b9\ua8ba\ua8be\ua8bf\ua8c0\ua8c1\ua8c2\ua8c3\ua8c4\ua8d1\ua8d2\ua8d3\ua8d8",
+        "punctuations": "!\u00a1'(),-.:;\u00bf? ",
+        "phonemes": null,
+        "is_unique": true,
+        "is_sorted": true
+    },
+    "add_blank": true,
+    "batch_group_size": 3,
+    "loss_masking": null,
+    "min_audio_len": 1,
+    "max_audio_len": Infinity,
+    "min_text_len": 1,
+    "max_text_len": Infinity,
+    "compute_f0": false,
+    "compute_energy": false,
+    "compute_linear_spec": true,
+    "precompute_num_workers": 0,
+    "start_by_longest": false,
+    "shuffle": false,
+    "drop_last": false,
+    "datasets": [
+        {
+            "formatter": "ljspeech",
+            "dataset_name": "",
+            "path": "./",
+            "meta_file_train": "sourashtra_dataset_female/processed/processed_female_ljspeech.csv",
+            "ignored_speakers": null,
+            "language": "",
+            "phonemizer": "",
+            "meta_file_val": "",
+            "meta_file_attn_mask": ""
+        }
+    ],
+    "test_sentences": [
+        [
+            "\ua88f\ua899\ua8a5\ua8b5\ua892\ua8c4 \ua88f\ua899\ua8a5\ua8c1 \ua8aa\ua8c1\ua8a5\ua8c4\ua8a5\ua8b8 \ua8b1\ua8c4\ua8a9\ua8b5\ua8ae\ua8c4 \ua899\ua8a5\ua8c4\ua8ad\ua8a5\ua8c4 \ua8af\ua8be\ua892\ua8c4\ua892\ua8b1\ua8c4\ua8a1\ua8c1 \ua8b2\ua8c1\ua8ab\ua8be \ua89a\ua8b8\ua894\ua8b8 \ua892\ua8b5\ua8ac\ua8a0\ua8b5\ua8a5\ua8b8\ua8aa\ua8c4 \ua88c\ua8ad\ua8c4\ua8ab\ua8c1 \ua88f\ua8a0\ua8c4\ua89c\ua8c4\ua8ab \ua892\ua8b5\ua8ac\ua8a0\ua8c1\ua892\ua8a5\ua8b6 \ua8b1\ua8bf \ua8aa\ua8be\ua8a5\ua8b6 \ua882\ua8aa\ua8b6 \ua892\ua8b3\ua8c0\ua8ad\ua8b6\ua8ac\ua8b6\ua8ab\ua8c1 \ua8a1\ua8be\ua8ad\ua8c4\ua8ab\ua892 \ua8b2\ua8b5\ua8ad\ua8c4 \ua897\ua8c1\ua892\ua8c4\ua892\ua89e\ua8b8\ua8b1\ua8c4",
+            "Sourashtra_Female_Sourashtra",
+            null,
+            "saz"
+        ]
+    ],
+    "eval_split_max_size": null,
+    "eval_split_size": 0.01,
+    "use_speaker_weighted_sampler": false,
+    "speaker_weighted_sampler_alpha": 1.0,
+    "use_language_weighted_sampler": false,
+    "language_weighted_sampler_alpha": 1.0,
+    "use_length_weighted_sampler": false,
+    "length_weighted_sampler_alpha": 1.0,
+    "lr_gen": 0.0002,
+    "lr_disc": 0.0002,
+    "lr_scheduler_gen": "ExponentialLR",
+    "lr_scheduler_gen_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "lr_scheduler_disc": "ExponentialLR",
+    "lr_scheduler_disc_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "kl_loss_alpha": 1.0,
+    "disc_loss_alpha": 1.0,
+    "gen_loss_alpha": 1.0,
+    "feat_loss_alpha": 1.0,
+    "mel_loss_alpha": 45.0,
+    "dur_loss_alpha": 1.0,
+    "speaker_encoder_loss_alpha": 1.0,
+    "return_wav": true,
+    "use_weighted_sampler": false,
+    "weighted_sampler_attrs": null,
+    "weighted_sampler_multipliers": null,
+    "r": 1,
+    "num_speakers": 0,
+    "use_speaker_embedding": false,
+    "speakers_file": null,
+    "speaker_embedding_channels": 256,
+    "language_ids_file": null,
+    "use_language_embedding": false,
+    "use_d_vector_file": false,
+    "d_vector_file": null,
+    "d_vector_dim": 0,
+    "github_branch": "inside_docker"
+}

Sourashtra-Female_Script-sourashtra/inference.py

@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+"""
+Synthesize Sourashtra-script text with the female VITS checkpoint.
+
+Usage:
+    python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav
+    python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" --gpu 1
+    python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" --cpu
+"""
+import argparse
+import os
+import re
+from pathlib import Path
+
+import torch
+from TTS.api import TTS
+
+CHECKPOINT = "best_model.pth"
+MODEL_DIR = Path(__file__).parent
+
+
+def clean_text(text: str) -> str:
+    text = re.sub(r"(?:\{[^}]*\}\s*)+", lambda m: re.findall(r"\{([^}]*)\}", m.group(0))[-1].strip(), text)
+    text = text.replace("꣎", "").replace("꣏", "")  # Sourashtra Danda / Double Danda
+    return re.sub(r"  +", " ", text).strip()
+
+
+def synthesize(tts, text: str, output_path: str) -> None:
+    tts.tts_to_file(text=text, file_path=output_path)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Female Sourashtra-script TTS inference")
+    parser.add_argument("text", help="Input text in Sourashtra script")
+    parser.add_argument("-o", "--output", default="output.wav", help="Output WAV path (default: output.wav)")
+    parser.add_argument("--gpu", type=int, default=0, help="GPU device ID (default: 0)")
+    parser.add_argument("--cpu", action="store_true", help="Force CPU inference")
+    args = parser.parse_args()
+
+    if not args.cpu:
+        os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
+
+    device = "cpu" if args.cpu else ("cuda:0" if torch.cuda.is_available() else "cpu")
+
+    tts = TTS(
+        model_path=str(MODEL_DIR / CHECKPOINT),
+        config_path=str(MODEL_DIR / "config.json"),
+        progress_bar=False,
+    )
+    tts.to(device)
+
+    text = clean_text(args.text)
+    print(f"Device    : {device}")
+    print(f"Text      : {text}")
+    print(f"Output    : {args.output}")
+
+    synthesize(tts, text, args.output)
+    print("Done.")
+
+
+if __name__ == "__main__":
+    main()

Sourashtra-Female_Script-sourashtra/requirements.txt

@@ -0,0 +1,4 @@
+coqui-tts
+torch
+soundfile
+numpy

Sourashtra-Female_Script-tamil/README.md

@@ -0,0 +1,46 @@
+---
+language:
+- saz
+- ta
+license: cc-by-4.0
+tags:
+- text-to-speech
+- tts
+- vits
+- sourashtra
+- tamil
+- low-resource
+pipeline_tag: text-to-speech
+---
+
+# VITS TTS — Sourashtra Female (Tamil Script)
+
+VITS text-to-speech model for the [Sourashtra language](https://en.wikipedia.org/wiki/Sourashtra_language), a minority language spoken primarily in Tamil Nadu, India. This model uses **Tamil script** (தமிழ்) input and synthesizes speech from a **female speaker**.
+
+## Model Details
+
+| Property | Value |
+|----------|-------|
+| Architecture | [VITS](https://arxiv.org/abs/2106.06103) (Coqui TTS) |
+| Speaker | Female |
+| Input script | Tamil |
+| Sample rate | 22050 Hz |
+| Training steps | 340,000 |
+
+## Setup
+
+```bash
+pip install -r requirements.txt
+```
+
+For GPU inference, install the CUDA-enabled PyTorch build matching your driver first — see [pytorch.org](https://pytorch.org/get-started/locally/).
+
+## Usage
+
+```bash
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav --gpu 1
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav --cpu
+```
+
+Input must be in Tamil script. For Sourashtra script input, use the `Sourashtra-Female_Script-sourashtra` model instead.

Sourashtra-Female_Script-tamil/best_model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd5c1c34d515f12307434249042c77e8c83a2c2a7c087b84127ea1ce7b78b911
+size 997811510

Sourashtra-Female_Script-tamil/config.json

@@ -0,0 +1,261 @@
+{
+    "output_path": "./output_Sourashtra_Female_tamil",
+    "logger_uri": null,
+    "run_name": "vits_Sourashtra_Female_tamil",
+    "project_name": null,
+    "run_description": "\ud83d\udc38Coqui trainer run.",
+    "print_step": 25,
+    "plot_step": 100,
+    "model_param_stats": false,
+    "wandb_entity": null,
+    "dashboard_logger": "tensorboard",
+    "save_on_interrupt": true,
+    "log_model_step": null,
+    "save_step": 10000,
+    "save_n_checkpoints": 7,
+    "save_checkpoints": true,
+    "save_all_best": false,
+    "save_best_after": 0,
+    "target_loss": null,
+    "print_eval": true,
+    "test_delay_epochs": -1,
+    "run_eval": true,
+    "run_eval_steps": null,
+    "distributed_backend": "nccl",
+    "distributed_url": "tcp://localhost:54321",
+    "mixed_precision": true,
+    "precision": "fp16",
+    "epochs": 500,
+    "batch_size": 16,
+    "eval_batch_size": 8,
+    "grad_clip": [
+        1000,
+        1000
+    ],
+    "scheduler_after_epoch": true,
+    "lr": 0.001,
+    "optimizer": "AdamW",
+    "optimizer_params": {
+        "betas": [
+            0.8,
+            0.99
+        ],
+        "eps": 1e-09,
+        "weight_decay": 0.01
+    },
+    "lr_scheduler": null,
+    "lr_scheduler_params": {},
+    "use_grad_scaler": false,
+    "allow_tf32": false,
+    "cudnn_enable": true,
+    "cudnn_deterministic": false,
+    "cudnn_benchmark": false,
+    "training_seed": 54321,
+    "model": "vits",
+    "num_loader_workers": 8,
+    "num_eval_loader_workers": 4,
+    "use_noise_augment": false,
+    "audio": {
+        "fft_size": 1024,
+        "sample_rate": 22050,
+        "win_length": 1024,
+        "hop_length": 256,
+        "num_mels": 80,
+        "mel_fmin": 0,
+        "mel_fmax": null
+    },
+    "model_args": {
+        "num_chars": 83,
+        "out_channels": 513,
+        "spec_segment_size": 32,
+        "hidden_channels": 192,
+        "hidden_channels_ffn_text_encoder": 768,
+        "num_heads_text_encoder": 2,
+        "num_layers_text_encoder": 6,
+        "kernel_size_text_encoder": 3,
+        "dropout_p_text_encoder": 0.1,
+        "dropout_p_duration_predictor": 0.5,
+        "kernel_size_posterior_encoder": 5,
+        "dilation_rate_posterior_encoder": 1,
+        "num_layers_posterior_encoder": 16,
+        "kernel_size_flow": 5,
+        "dilation_rate_flow": 1,
+        "num_layers_flow": 4,
+        "resblock_type_decoder": "1",
+        "resblock_kernel_sizes_decoder": [
+            3,
+            7,
+            11
+        ],
+        "resblock_dilation_sizes_decoder": [
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ]
+        ],
+        "upsample_rates_decoder": [
+            8,
+            8,
+            2,
+            2
+        ],
+        "upsample_initial_channel_decoder": 512,
+        "upsample_kernel_sizes_decoder": [
+            16,
+            16,
+            4,
+            4
+        ],
+        "periods_multi_period_discriminator": [
+            2,
+            3,
+            5,
+            7,
+            11
+        ],
+        "use_sdp": true,
+        "noise_scale": 1.0,
+        "inference_noise_scale": 0.667,
+        "length_scale": 1,
+        "noise_scale_dp": 1.0,
+        "inference_noise_scale_dp": 1.0,
+        "max_inference_len": null,
+        "init_discriminator": true,
+        "use_spectral_norm_disriminator": false,
+        "use_speaker_embedding": false,
+        "num_speakers": 0,
+        "speakers_file": null,
+        "d_vector_file": null,
+        "speaker_embedding_channels": 256,
+        "use_d_vector_file": false,
+        "d_vector_dim": 0,
+        "detach_dp_input": true,
+        "use_language_embedding": false,
+        "embedded_language_dim": 4,
+        "num_languages": 0,
+        "language_ids_file": null,
+        "use_speaker_encoder_as_loss": false,
+        "speaker_encoder_config_path": "",
+        "speaker_encoder_model_path": "",
+        "condition_dp_on_speaker": true,
+        "freeze_encoder": false,
+        "freeze_DP": false,
+        "freeze_PE": false,
+        "freeze_flow_decoder": false,
+        "freeze_waveform_decoder": false,
+        "encoder_sample_rate": null,
+        "interpolate_z": true,
+        "reinit_DP": false,
+        "reinit_text_encoder": false
+    },
+    "_supports_cloning": false,
+    "use_phonemes": false,
+    "phonemizer": null,
+    "phoneme_language": "en-us",
+    "compute_input_seq_cache": true,
+    "text_cleaner": "multilingual_cleaners",
+    "enable_eos_bos_chars": false,
+    "test_sentences_file": "",
+    "phoneme_cache_path": "./output_Sourashtra_Female_tamil/phoneme_cache",
+    "characters": {
+        "characters_class": "TTS.tts.models.vits.VitsCharacters",
+        "vocab_dict": null,
+        "pad": "<PAD>",
+        "eos": "<EOS>",
+        "bos": "<BOS>",
+        "blank": "<BLNK>",
+        "characters": " !(),-/1234?\u00b2\u00b3\u02bc\u0b83\u0b85\u0b86\u0b87\u0b88\u0b89\u0b8a\u0b8e\u0b8f\u0b90\u0b92\u0b93\u0b94\u0b95\u0b99\u0b9a\u0b9c\u0b9e\u0b9f\u0ba3\u0ba4\u0ba8\u0ba9\u0baa\u0bae\u0baf\u0bb0\u0bb1\u0bb2\u0bb3\u0bb4\u0bb5\u0bb6\u0bb7\u0bb8\u0bb9\u0bbe\u0bbf\u0bc0\u0bc1\u0bc2\u0bc6\u0bc7\u0bc8\u0bca\u0bcb\u0bcc\u0bcd\u0be7\u0bee\u200c\u2074\ua8d2",
+        "punctuations": "!\u00a1'(),-.:;\u00bf? ",
+        "phonemes": null,
+        "is_unique": true,
+        "is_sorted": true
+    },
+    "add_blank": true,
+    "batch_group_size": 3,
+    "loss_masking": null,
+    "min_audio_len": 1,
+    "max_audio_len": Infinity,
+    "min_text_len": 1,
+    "max_text_len": Infinity,
+    "compute_f0": false,
+    "compute_energy": false,
+    "compute_linear_spec": true,
+    "precompute_num_workers": 0,
+    "start_by_longest": false,
+    "shuffle": false,
+    "drop_last": false,
+    "datasets": [
+        {
+            "formatter": "ljspeech",
+            "dataset_name": "",
+            "path": "./",
+            "meta_file_train": "sourashtra_dataset_female/processed/processed_female_tamil_ljspeech.csv",
+            "ignored_speakers": null,
+            "language": "",
+            "phonemizer": "",
+            "meta_file_val": "",
+            "meta_file_attn_mask": ""
+        }
+    ],
+    "test_sentences": [
+        [
+            "\u0b92\u0b9c\u0ba9\u0bbe\u0b95\u0bcd \u0b92\u0b9c\u0ba9\u0bca \u0bae\u0bca\u0ba9\u0bcd\u0ba9\u0bc1 \u0bb8\u0bcd\u0baa\u0bbe4\u0bb5\u0bcd \u0b9c\u0ba9\u0bcd\u0bb2\u0ba9\u0bcd \u0bb6\u0bc6\u0b95\u0bcd\u0b95\u0bb8\u0bcd\u0ba4\u0bca \u0bb9\u0bca\u0baf\u0bc6 \u0b9c\u0bc12\u0b95\u0bc13 \u0b95\u0bbe\u0bb0\u0ba3\u0bbe\u0ba9\u0bc1\u0bae\u0bcd \u0b8e\u0bb2\u0bcd\u0baf\u0bca \u0b92\u0ba3\u0bcd\u0b9f\u0bcd\u0baf \u0b95\u0bbe\u0bb0\u0ba3\u0bca\u0b95\u0ba9\u0bbf \u0bb8\u0bc7 \u0bae\u0bc6\u0ba9\u0bbf \u0b85\u0bae\u0bbf \u0b95\u0bb3\u0bc8\u0bb2\u0bbf\u0bb0\u0bbf\u0baf\u0bca \u0ba4\u0bc6\u0bb2\u0bcd\u0baf\u0b95 \u0bb9\u0bbe\u0bb2\u0bcd \u0b9a\u0bca\u0b95\u0bcd\u0b95\u0b9f\u0bc13\u0bb8\u0bcd",
+            "Sourashtra_Female_Tamil",
+            null,
+            "ta"
+        ]
+    ],
+    "eval_split_max_size": null,
+    "eval_split_size": 0.01,
+    "use_speaker_weighted_sampler": false,
+    "speaker_weighted_sampler_alpha": 1.0,
+    "use_language_weighted_sampler": false,
+    "language_weighted_sampler_alpha": 1.0,
+    "use_length_weighted_sampler": false,
+    "length_weighted_sampler_alpha": 1.0,
+    "lr_gen": 0.0002,
+    "lr_disc": 0.0002,
+    "lr_scheduler_gen": "ExponentialLR",
+    "lr_scheduler_gen_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "lr_scheduler_disc": "ExponentialLR",
+    "lr_scheduler_disc_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "kl_loss_alpha": 1.0,
+    "disc_loss_alpha": 1.0,
+    "gen_loss_alpha": 1.0,
+    "feat_loss_alpha": 1.0,
+    "mel_loss_alpha": 45.0,
+    "dur_loss_alpha": 1.0,
+    "speaker_encoder_loss_alpha": 1.0,
+    "return_wav": true,
+    "use_weighted_sampler": false,
+    "weighted_sampler_attrs": {},
+    "weighted_sampler_multipliers": {},
+    "r": 1,
+    "num_speakers": 0,
+    "use_speaker_embedding": false,
+    "speakers_file": null,
+    "speaker_embedding_channels": 256,
+    "language_ids_file": null,
+    "use_language_embedding": false,
+    "use_d_vector_file": false,
+    "d_vector_file": null,
+    "d_vector_dim": 0,
+    "github_branch": "inside_docker"
+}

Sourashtra-Female_Script-tamil/inference.py

@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+"""
+Synthesize Tamil-script text with the female VITS checkpoint.
+
+Usage:
+    python inference.py "சொராஷ்ட்ர மொழி" -o output.wav
+    python inference.py "சொராஷ்ட்ர மொழி" --gpu 1
+    python inference.py "சொராஷ்ட்ர மொழி" --cpu
+"""
+import argparse
+import os
+import re
+import unicodedata
+from pathlib import Path
+
+import torch
+from TTS.api import TTS
+
+CHECKPOINT = "best_model.pth"
+MODEL_DIR = Path(__file__).parent
+
+
+def clean_text(text: str) -> str:
+    text = re.sub(r"(?:\{[^}]*\}\s*)+", lambda m: re.findall(r"\{([^}]*)\}", m.group(0))[-1].strip(), text)
+    text = unicodedata.normalize("NFC", text)
+    text = text.replace(":", "").replace(".", "").replace("'", "").replace("’", "")
+    return re.sub(r"  +", " ", text).strip()
+
+
+def synthesize(tts, text: str, output_path: str) -> None:
+    tts.tts_to_file(text=text, file_path=output_path)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Female Tamil-script TTS inference")
+    parser.add_argument("text", help="Input text in Tamil script")
+    parser.add_argument("-o", "--output", default="output.wav", help="Output WAV path (default: output.wav)")
+    parser.add_argument("--gpu", type=int, default=0, help="GPU device ID (default: 0)")
+    parser.add_argument("--cpu", action="store_true", help="Force CPU inference")
+    args = parser.parse_args()
+
+    if not args.cpu:
+        os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
+
+    device = "cpu" if args.cpu else ("cuda:0" if torch.cuda.is_available() else "cpu")
+
+    tts = TTS(
+        model_path=str(MODEL_DIR / CHECKPOINT),
+        config_path=str(MODEL_DIR / "config.json"),
+        progress_bar=False,
+    )
+    tts.to(device)
+
+    text = clean_text(args.text)
+    print(f"Device    : {device}")
+    print(f"Text      : {text}")
+    print(f"Output    : {args.output}")
+
+    synthesize(tts, text, args.output)
+    print("Done.")
+
+
+if __name__ == "__main__":
+    main()

Sourashtra-Female_Script-tamil/requirements.txt

@@ -0,0 +1,4 @@
+coqui-tts
+torch
+soundfile
+numpy

Sourashtra-Male_Script-sourashtra/README.md

@@ -0,0 +1,44 @@
+---
+language:
+- saz
+license: cc-by-4.0
+tags:
+- text-to-speech
+- tts
+- vits
+- sourashtra
+- low-resource
+pipeline_tag: text-to-speech
+---
+
+# VITS TTS — Sourashtra Male (Sourashtra Script)
+
+VITS text-to-speech model for the [Sourashtra language](https://en.wikipedia.org/wiki/Sourashtra_language), a minority language spoken primarily in Tamil Nadu, India. This model uses **Sourashtra script** (ꢪꢾꢥꢶꢒ) input and synthesizes speech from a **male speaker**.
+
+## Model Details
+
+| Property | Value |
+|----------|-------|
+| Architecture | [VITS](https://arxiv.org/abs/2106.06103) (Coqui TTS) |
+| Speaker | Male |
+| Input script | Sourashtra (Unicode block U+A880–U+A8D9) |
+| Sample rate | 22050 Hz |
+| Training steps | 300,000 |
+
+## Setup
+
+```bash
+pip install -r requirements.txt
+```
+
+For GPU inference, install the CUDA-enabled PyTorch build matching your driver first — see [pytorch.org](https://pytorch.org/get-started/locally/).
+
+## Usage
+
+```bash
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav --gpu 1
+python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav --cpu
+```
+
+Input must be in Sourashtra script. For Tamil script input, use the `Sourashtra-Male_Script-tamil` model instead.

Sourashtra-Male_Script-sourashtra/best_model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:820d5f9b3d55ad5331319a692f34af87c9721ed53802bdfc17dd32fb9612ac9e
+size 997857718

Sourashtra-Male_Script-sourashtra/config.json

@@ -0,0 +1,261 @@
+{
+    "output_path": "./output_Sourashtra_Male_Sourashtra",
+    "logger_uri": null,
+    "run_name": "vits_Sourashtra_Male_Sourashtra_sourashtra",
+    "project_name": null,
+    "run_description": "\ud83d\udc38Coqui trainer run.",
+    "print_step": 25,
+    "plot_step": 100,
+    "model_param_stats": false,
+    "wandb_entity": null,
+    "dashboard_logger": "tensorboard",
+    "save_on_interrupt": true,
+    "log_model_step": null,
+    "save_step": 10000,
+    "save_n_checkpoints": 7,
+    "save_checkpoints": true,
+    "save_all_best": false,
+    "save_best_after": 0,
+    "target_loss": null,
+    "print_eval": true,
+    "test_delay_epochs": -1,
+    "run_eval": true,
+    "run_eval_steps": null,
+    "distributed_backend": "nccl",
+    "distributed_url": "tcp://localhost:54321",
+    "mixed_precision": true,
+    "precision": "fp16",
+    "epochs": 500,
+    "batch_size": 16,
+    "eval_batch_size": 8,
+    "grad_clip": [
+        1000,
+        1000
+    ],
+    "scheduler_after_epoch": true,
+    "lr": 0.001,
+    "optimizer": "AdamW",
+    "optimizer_params": {
+        "betas": [
+            0.8,
+            0.99
+        ],
+        "eps": 1e-09,
+        "weight_decay": 0.01
+    },
+    "lr_scheduler": null,
+    "lr_scheduler_params": null,
+    "use_grad_scaler": false,
+    "allow_tf32": false,
+    "cudnn_enable": true,
+    "cudnn_deterministic": false,
+    "cudnn_benchmark": false,
+    "training_seed": 54321,
+    "model": "vits",
+    "num_loader_workers": 8,
+    "num_eval_loader_workers": 4,
+    "use_noise_augment": false,
+    "audio": {
+        "fft_size": 1024,
+        "sample_rate": 22050,
+        "win_length": 1024,
+        "hop_length": 256,
+        "num_mels": 80,
+        "mel_fmin": 0,
+        "mel_fmax": null
+    },
+    "model_args": {
+        "num_chars": 103,
+        "out_channels": 513,
+        "spec_segment_size": 32,
+        "hidden_channels": 192,
+        "hidden_channels_ffn_text_encoder": 768,
+        "num_heads_text_encoder": 2,
+        "num_layers_text_encoder": 6,
+        "kernel_size_text_encoder": 3,
+        "dropout_p_text_encoder": 0.1,
+        "dropout_p_duration_predictor": 0.5,
+        "kernel_size_posterior_encoder": 5,
+        "dilation_rate_posterior_encoder": 1,
+        "num_layers_posterior_encoder": 16,
+        "kernel_size_flow": 5,
+        "dilation_rate_flow": 1,
+        "num_layers_flow": 4,
+        "resblock_type_decoder": "1",
+        "resblock_kernel_sizes_decoder": [
+            3,
+            7,
+            11
+        ],
+        "resblock_dilation_sizes_decoder": [
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ]
+        ],
+        "upsample_rates_decoder": [
+            8,
+            8,
+            2,
+            2
+        ],
+        "upsample_initial_channel_decoder": 512,
+        "upsample_kernel_sizes_decoder": [
+            16,
+            16,
+            4,
+            4
+        ],
+        "periods_multi_period_discriminator": [
+            2,
+            3,
+            5,
+            7,
+            11
+        ],
+        "use_sdp": true,
+        "noise_scale": 1.0,
+        "inference_noise_scale": 0.667,
+        "length_scale": 1,
+        "noise_scale_dp": 1.0,
+        "inference_noise_scale_dp": 1.0,
+        "max_inference_len": null,
+        "init_discriminator": true,
+        "use_spectral_norm_disriminator": false,
+        "use_speaker_embedding": false,
+        "num_speakers": 0,
+        "speakers_file": null,
+        "d_vector_file": null,
+        "speaker_embedding_channels": 256,
+        "use_d_vector_file": false,
+        "d_vector_dim": 0,
+        "detach_dp_input": true,
+        "use_language_embedding": false,
+        "embedded_language_dim": 4,
+        "num_languages": 0,
+        "language_ids_file": null,
+        "use_speaker_encoder_as_loss": false,
+        "speaker_encoder_config_path": "",
+        "speaker_encoder_model_path": "",
+        "condition_dp_on_speaker": true,
+        "freeze_encoder": false,
+        "freeze_DP": false,
+        "freeze_PE": false,
+        "freeze_flow_decoder": false,
+        "freeze_waveform_decoder": false,
+        "encoder_sample_rate": null,
+        "interpolate_z": true,
+        "reinit_DP": false,
+        "reinit_text_encoder": false
+    },
+    "_supports_cloning": false,
+    "use_phonemes": false,
+    "phonemizer": null,
+    "phoneme_language": "en-us",
+    "compute_input_seq_cache": true,
+    "text_cleaner": "multilingual_cleaners",
+    "enable_eos_bos_chars": false,
+    "test_sentences_file": "",
+    "phoneme_cache_path": "./output_Sourashtra_Male_Sourashtra/phoneme_cache",
+    "characters": {
+        "characters_class": "TTS.tts.models.vits.VitsCharacters",
+        "vocab_dict": null,
+        "pad": "<PAD>",
+        "eos": "<EOS>",
+        "bos": "<BOS>",
+        "blank": "<BLNK>",
+        "characters": " !'(),-.1237:?ht\u00b3\u200d\u2074\ua882\ua883\ua884\ua885\ua886\ua887\ua888\ua889\ua88c\ua88d\ua88e\ua88f\ua890\ua891\ua892\ua893\ua894\ua895\ua896\ua897\ua898\ua899\ua89a\ua89b\ua89c\ua89d\ua89e\ua89f\ua8a0\ua8a1\ua8a2\ua8a3\ua8a4\ua8a5\ua8a6\ua8a7\ua8a8\ua8a9\ua8aa\ua8ab\ua8ac\ua8ad\ua8ae\ua8af\ua8b0\ua8b1\ua8b2\ua8b3\ua8b4\ua8b5\ua8b6\ua8b7\ua8b8\ua8b9\ua8ba\ua8be\ua8bf\ua8c0\ua8c1\ua8c2\ua8c3\ua8c4\ua8ce\ua8cf\ua8d1\ua8d2\ua8d3\ua8d5\ua8d7",
+        "punctuations": "!\u00a1'(),-.:;\u00bf? ",
+        "phonemes": null,
+        "is_unique": true,
+        "is_sorted": true
+    },
+    "add_blank": true,
+    "batch_group_size": 3,
+    "loss_masking": null,
+    "min_audio_len": 1,
+    "max_audio_len": Infinity,
+    "min_text_len": 1,
+    "max_text_len": Infinity,
+    "compute_f0": false,
+    "compute_energy": false,
+    "compute_linear_spec": true,
+    "precompute_num_workers": 0,
+    "start_by_longest": false,
+    "shuffle": false,
+    "drop_last": false,
+    "datasets": [
+        {
+            "formatter": "ljspeech",
+            "dataset_name": "",
+            "path": "./",
+            "meta_file_train": "sourashtra_dataset/processed/processed_sourashtra_ljspeech.csv",
+            "ignored_speakers": null,
+            "language": "",
+            "phonemizer": "",
+            "meta_file_val": "",
+            "meta_file_attn_mask": ""
+        }
+    ],
+    "test_sentences": [
+        [
+            "\ua88f\ua899\ua8a5\ua8b5\ua892\ua8c4 \ua88f\ua899\ua8a5\ua8c1 \ua8aa\ua8c1\ua8a5\ua8c4\ua8a5\ua8b8 \ua8b1\ua8c4\ua8a9\ua8b5\ua8ae\ua8c4 \ua899\ua8a5\ua8c4\ua8ad\ua8a5\ua8c4 \ua8af\ua8be\ua892\ua8c4\ua892\ua8b1\ua8c4\ua8a1\ua8c1 \ua8b2\ua8c1\ua8ab\ua8be \ua89a\ua8b8\ua894\ua8b8 \ua892\ua8b5\ua8ac\ua8a0\ua8b5\ua8a5\ua8b8\ua8aa\ua8c4 \ua88c\ua8ad\ua8c4\ua8ab\ua8c1 \ua88f\ua8a0\ua8c4\ua89c\ua8c4\ua8ab \ua892\ua8b5\ua8ac\ua8a0\ua8c1\ua892\ua8a5\ua8b6 \ua8b1\ua8bf \ua8aa\ua8be\ua8a5\ua8b6 \ua882\ua8aa\ua8b6 \ua892\ua8b3\ua8c0\ua8ad\ua8b6\ua8ac\ua8b6\ua8ab\ua8c1 \ua8a1\ua8be\ua8ad\ua8c4\ua8ab\ua892 \ua8b2\ua8b5\ua8ad\ua8c4 \ua897\ua8c1\ua892\ua8c4\ua892\ua89e\ua8b8\ua8b1\ua8c4",
+            "Sourashtra_Male_Sourashtra",
+            null,
+            "saz"
+        ]
+    ],
+    "eval_split_max_size": null,
+    "eval_split_size": 0.01,
+    "use_speaker_weighted_sampler": false,
+    "speaker_weighted_sampler_alpha": 1.0,
+    "use_language_weighted_sampler": false,
+    "language_weighted_sampler_alpha": 1.0,
+    "use_length_weighted_sampler": false,
+    "length_weighted_sampler_alpha": 1.0,
+    "lr_gen": 0.0002,
+    "lr_disc": 0.0002,
+    "lr_scheduler_gen": "ExponentialLR",
+    "lr_scheduler_gen_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "lr_scheduler_disc": "ExponentialLR",
+    "lr_scheduler_disc_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "kl_loss_alpha": 1.0,
+    "disc_loss_alpha": 1.0,
+    "gen_loss_alpha": 1.0,
+    "feat_loss_alpha": 1.0,
+    "mel_loss_alpha": 45.0,
+    "dur_loss_alpha": 1.0,
+    "speaker_encoder_loss_alpha": 1.0,
+    "return_wav": true,
+    "use_weighted_sampler": false,
+    "weighted_sampler_attrs": null,
+    "weighted_sampler_multipliers": null,
+    "r": 1,
+    "num_speakers": 0,
+    "use_speaker_embedding": false,
+    "speakers_file": null,
+    "speaker_embedding_channels": 256,
+    "language_ids_file": null,
+    "use_language_embedding": false,
+    "use_d_vector_file": false,
+    "d_vector_file": null,
+    "d_vector_dim": 0,
+    "github_branch": "inside_docker"
+}

Sourashtra-Male_Script-sourashtra/inference.py

@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+"""
+Synthesize Sourashtra-script text with the male VITS checkpoint.
+
+Usage:
+    python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" -o output.wav
+    python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" --gpu 1
+    python inference.py "ꢪꢾꢥꢶꢒ ꢪꢒꢡ" --cpu
+"""
+import argparse
+import os
+import re
+from pathlib import Path
+
+import torch
+from TTS.api import TTS
+
+CHECKPOINT = "best_model.pth"
+MODEL_DIR = Path(__file__).parent
+
+
+def clean_text(text: str) -> str:
+    text = re.sub(r"(?:\{[^}]*\}\s*)+", lambda m: re.findall(r"\{([^}]*)\}", m.group(0))[-1].strip(), text)
+    text = text.replace("꣎", "").replace("꣏", "")  # Sourashtra Danda / Double Danda
+    return re.sub(r"  +", " ", text).strip()
+
+
+def synthesize(tts, text: str, output_path: str) -> None:
+    tts.tts_to_file(text=text, file_path=output_path)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Male Sourashtra-script TTS inference")
+    parser.add_argument("text", help="Input text in Sourashtra script")
+    parser.add_argument("-o", "--output", default="output.wav", help="Output WAV path (default: output.wav)")
+    parser.add_argument("--gpu", type=int, default=0, help="GPU device ID (default: 0)")
+    parser.add_argument("--cpu", action="store_true", help="Force CPU inference")
+    args = parser.parse_args()
+
+    if not args.cpu:
+        os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
+
+    device = "cpu" if args.cpu else ("cuda:0" if torch.cuda.is_available() else "cpu")
+
+    tts = TTS(
+        model_path=str(MODEL_DIR / CHECKPOINT),
+        config_path=str(MODEL_DIR / "config.json"),
+        progress_bar=False,
+    )
+    tts.to(device)
+
+    text = clean_text(args.text)
+    print(f"Device    : {device}")
+    print(f"Text      : {text}")
+    print(f"Output    : {args.output}")
+
+    synthesize(tts, text, args.output)
+    print("Done.")
+
+
+if __name__ == "__main__":
+    main()

Sourashtra-Male_Script-sourashtra/requirements.txt

@@ -0,0 +1,4 @@
+coqui-tts
+torch
+soundfile
+numpy

Sourashtra-Male_Script-tamil/README.md

@@ -0,0 +1,46 @@
+---
+language:
+- saz
+- ta
+license: cc-by-4.0
+tags:
+- text-to-speech
+- tts
+- vits
+- sourashtra
+- tamil
+- low-resource
+pipeline_tag: text-to-speech
+---
+
+# VITS TTS — Sourashtra Male (Tamil Script)
+
+VITS text-to-speech model for the [Sourashtra language](https://en.wikipedia.org/wiki/Sourashtra_language), a minority language spoken primarily in Tamil Nadu, India. This model uses **Tamil script** (தமிழ்) input and synthesizes speech from a **male speaker**.
+
+## Model Details
+
+| Property | Value |
+|----------|-------|
+| Architecture | [VITS](https://arxiv.org/abs/2106.06103) (Coqui TTS) |
+| Speaker | Male |
+| Input script | Tamil |
+| Sample rate | 22050 Hz |
+| Training steps | 300,000 |
+
+## Setup
+
+```bash
+pip install -r requirements.txt
+```
+
+For GPU inference, install the CUDA-enabled PyTorch build matching your driver first — see [pytorch.org](https://pytorch.org/get-started/locally/).
+
+## Usage
+
+```bash
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav --gpu 1
+python inference.py "சொராஷ்ட்ர மொழி" -o output.wav --cpu
+```
+
+Input must be in Tamil script. For Sourashtra script input, use the `Sourashtra-Male_Script-sourashtra` model instead.

Sourashtra-Male_Script-tamil/best_model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f93412aa19745844cda1675c5e33357ce0de0267c066756ae3a57a8f073d40e0
+size 997809206

Sourashtra-Male_Script-tamil/config.json

@@ -0,0 +1,261 @@
+{
+    "output_path": "./output_Sourashtra_Male_Tamil",
+    "logger_uri": null,
+    "run_name": "vits_Sourashtra_Male_Tamil_tamil",
+    "project_name": null,
+    "run_description": "\ud83d\udc38Coqui trainer run.",
+    "print_step": 25,
+    "plot_step": 100,
+    "model_param_stats": false,
+    "wandb_entity": null,
+    "dashboard_logger": "tensorboard",
+    "save_on_interrupt": true,
+    "log_model_step": null,
+    "save_step": 10000,
+    "save_n_checkpoints": 7,
+    "save_checkpoints": true,
+    "save_all_best": false,
+    "save_best_after": 0,
+    "target_loss": null,
+    "print_eval": true,
+    "test_delay_epochs": -1,
+    "run_eval": true,
+    "run_eval_steps": null,
+    "distributed_backend": "nccl",
+    "distributed_url": "tcp://localhost:54321",
+    "mixed_precision": true,
+    "precision": "fp16",
+    "epochs": 500,
+    "batch_size": 16,
+    "eval_batch_size": 8,
+    "grad_clip": [
+        1000,
+        1000
+    ],
+    "scheduler_after_epoch": true,
+    "lr": 0.001,
+    "optimizer": "AdamW",
+    "optimizer_params": {
+        "betas": [
+            0.8,
+            0.99
+        ],
+        "eps": 1e-09,
+        "weight_decay": 0.01
+    },
+    "lr_scheduler": null,
+    "lr_scheduler_params": {},
+    "use_grad_scaler": false,
+    "allow_tf32": false,
+    "cudnn_enable": true,
+    "cudnn_deterministic": false,
+    "cudnn_benchmark": false,
+    "training_seed": 54321,
+    "model": "vits",
+    "num_loader_workers": 8,
+    "num_eval_loader_workers": 4,
+    "use_noise_augment": false,
+    "audio": {
+        "fft_size": 1024,
+        "sample_rate": 22050,
+        "win_length": 1024,
+        "hop_length": 256,
+        "num_mels": 80,
+        "mel_fmin": 0,
+        "mel_fmax": null
+    },
+    "model_args": {
+        "num_chars": 82,
+        "out_channels": 513,
+        "spec_segment_size": 32,
+        "hidden_channels": 192,
+        "hidden_channels_ffn_text_encoder": 768,
+        "num_heads_text_encoder": 2,
+        "num_layers_text_encoder": 6,
+        "kernel_size_text_encoder": 3,
+        "dropout_p_text_encoder": 0.1,
+        "dropout_p_duration_predictor": 0.5,
+        "kernel_size_posterior_encoder": 5,
+        "dilation_rate_posterior_encoder": 1,
+        "num_layers_posterior_encoder": 16,
+        "kernel_size_flow": 5,
+        "dilation_rate_flow": 1,
+        "num_layers_flow": 4,
+        "resblock_type_decoder": "1",
+        "resblock_kernel_sizes_decoder": [
+            3,
+            7,
+            11
+        ],
+        "resblock_dilation_sizes_decoder": [
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ],
+            [
+                1,
+                3,
+                5
+            ]
+        ],
+        "upsample_rates_decoder": [
+            8,
+            8,
+            2,
+            2
+        ],
+        "upsample_initial_channel_decoder": 512,
+        "upsample_kernel_sizes_decoder": [
+            16,
+            16,
+            4,
+            4
+        ],
+        "periods_multi_period_discriminator": [
+            2,
+            3,
+            5,
+            7,
+            11
+        ],
+        "use_sdp": true,
+        "noise_scale": 1.0,
+        "inference_noise_scale": 0.667,
+        "length_scale": 1,
+        "noise_scale_dp": 1.0,
+        "inference_noise_scale_dp": 1.0,
+        "max_inference_len": null,
+        "init_discriminator": true,
+        "use_spectral_norm_disriminator": false,
+        "use_speaker_embedding": false,
+        "num_speakers": 0,
+        "speakers_file": null,
+        "d_vector_file": null,
+        "speaker_embedding_channels": 256,
+        "use_d_vector_file": false,
+        "d_vector_dim": 0,
+        "detach_dp_input": true,
+        "use_language_embedding": false,
+        "embedded_language_dim": 4,
+        "num_languages": 0,
+        "language_ids_file": null,
+        "use_speaker_encoder_as_loss": false,
+        "speaker_encoder_config_path": "",
+        "speaker_encoder_model_path": "",
+        "condition_dp_on_speaker": true,
+        "freeze_encoder": false,
+        "freeze_DP": false,
+        "freeze_PE": false,
+        "freeze_flow_decoder": false,
+        "freeze_waveform_decoder": false,
+        "encoder_sample_rate": null,
+        "interpolate_z": true,
+        "reinit_DP": false,
+        "reinit_text_encoder": false
+    },
+    "_supports_cloning": false,
+    "use_phonemes": false,
+    "phonemizer": null,
+    "phoneme_language": "en-us",
+    "compute_input_seq_cache": true,
+    "text_cleaner": "multilingual_cleaners",
+    "enable_eos_bos_chars": false,
+    "test_sentences_file": "",
+    "phoneme_cache_path": "./output_Sourashtra_Male_Tamil/phoneme_cache",
+    "characters": {
+        "characters_class": "TTS.tts.models.vits.VitsCharacters",
+        "vocab_dict": null,
+        "pad": "<PAD>",
+        "eos": "<EOS>",
+        "bos": "<BOS>",
+        "blank": "<BLNK>",
+        "characters": " !(),-12345?ht\u00b2\u00b3\u02bc\u0b83\u0b85\u0b86\u0b87\u0b88\u0b89\u0b8a\u0b8e\u0b8f\u0b90\u0b92\u0b93\u0b94\u0b95\u0b99\u0b9a\u0b9c\u0b9e\u0b9f\u0ba3\u0ba4\u0ba8\u0ba9\u0baa\u0bae\u0baf\u0bb0\u0bb1\u0bb2\u0bb3\u0bb4\u0bb5\u0bb6\u0bb7\u0bb8\u0bb9\u0bbe\u0bbf\u0bc0\u0bc1\u0bc2\u0bc6\u0bc7\u0bc8\u0bca\u0bcb\u0bcc\u0bcd\u200c\u2074",
+        "punctuations": "!\u00a1'(),-.:;\u00bf? ",
+        "phonemes": null,
+        "is_unique": true,
+        "is_sorted": true
+    },
+    "add_blank": true,
+    "batch_group_size": 3,
+    "loss_masking": null,
+    "min_audio_len": 1,
+    "max_audio_len": Infinity,
+    "min_text_len": 1,
+    "max_text_len": Infinity,
+    "compute_f0": false,
+    "compute_energy": false,
+    "compute_linear_spec": true,
+    "precompute_num_workers": 0,
+    "start_by_longest": false,
+    "shuffle": false,
+    "drop_last": false,
+    "datasets": [
+        {
+            "formatter": "ljspeech",
+            "dataset_name": "",
+            "path": "./",
+            "meta_file_train": "sourashtra_dataset/processed/processed_sourashtra_tamil_ljspeech.csv",
+            "ignored_speakers": null,
+            "language": "",
+            "phonemizer": "",
+            "meta_file_val": "",
+            "meta_file_attn_mask": ""
+        }
+    ],
+    "test_sentences": [
+        [
+            "\u0b92\u0b9c\u0ba9\u0bbe\u0b95\u0bcd \u0b92\u0b9c\u0ba9\u0bca \u0bae\u0bca\u0ba9\u0bcd\u0ba9\u0bc1 \u0bb8\u0bcd\u0baa\u0bbe4\u0bb5\u0bcd \u0b9c\u0ba9\u0bcd\u0bb2\u0ba9\u0bcd \u0bb6\u0bc6\u0b95\u0bcd\u0b95\u0bb8\u0bcd\u0ba4\u0bca \u0bb9\u0bca\u0baf\u0bc6 \u0b9c\u0bc12\u0b95\u0bc13 \u0b95\u0bbe\u0bb0\u0ba3\u0bbe\u0ba9\u0bc1\u0bae\u0bcd \u0b8e\u0bb2\u0bcd\u0baf\u0bca \u0b92\u0ba3\u0bcd\u0b9f\u0bcd\u0baf \u0b95\u0bbe\u0bb0\u0ba3\u0bca\u0b95\u0ba9\u0bbf \u0bb8\u0bc7 \u0bae\u0bc6\u0ba9\u0bbf \u0b85\u0bae\u0bbf \u0b95\u0bb3\u0bc8\u0bb2\u0bbf\u0bb0\u0bbf\u0baf\u0bca \u0ba4\u0bc6\u0bb2\u0bcd\u0baf\u0b95 \u0bb9\u0bbe\u0bb2\u0bcd \u0b9a\u0bca\u0b95\u0bcd\u0b95\u0b9f\u0bc13\u0bb8\u0bcd",
+            "Sourashtra_Male_Tamil",
+            null,
+            "ta"
+        ]
+    ],
+    "eval_split_max_size": null,
+    "eval_split_size": 0.01,
+    "use_speaker_weighted_sampler": false,
+    "speaker_weighted_sampler_alpha": 1.0,
+    "use_language_weighted_sampler": false,
+    "language_weighted_sampler_alpha": 1.0,
+    "use_length_weighted_sampler": false,
+    "length_weighted_sampler_alpha": 1.0,
+    "lr_gen": 0.0002,
+    "lr_disc": 0.0002,
+    "lr_scheduler_gen": "ExponentialLR",
+    "lr_scheduler_gen_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "lr_scheduler_disc": "ExponentialLR",
+    "lr_scheduler_disc_params": {
+        "gamma": 0.999875,
+        "last_epoch": -1
+    },
+    "kl_loss_alpha": 1.0,
+    "disc_loss_alpha": 1.0,
+    "gen_loss_alpha": 1.0,
+    "feat_loss_alpha": 1.0,
+    "mel_loss_alpha": 45.0,
+    "dur_loss_alpha": 1.0,
+    "speaker_encoder_loss_alpha": 1.0,
+    "return_wav": true,
+    "use_weighted_sampler": false,
+    "weighted_sampler_attrs": {},
+    "weighted_sampler_multipliers": {},
+    "r": 1,
+    "num_speakers": 0,
+    "use_speaker_embedding": false,
+    "speakers_file": null,
+    "speaker_embedding_channels": 256,
+    "language_ids_file": null,
+    "use_language_embedding": false,
+    "use_d_vector_file": false,
+    "d_vector_file": null,
+    "d_vector_dim": 0,
+    "github_branch": "inside_docker"
+}

Sourashtra-Male_Script-tamil/inference.py

@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+"""
+Synthesize Tamil-script text with the male VITS checkpoint.
+
+Usage:
+    python inference.py "சொராஷ்ட்ர மொழி" -o output.wav
+    python inference.py "சொராஷ்ட்ர மொழி" --gpu 1
+    python inference.py "சொராஷ்ட்ர மொழி" --cpu
+"""
+import argparse
+import os
+import re
+import unicodedata
+from pathlib import Path
+
+import torch
+from TTS.api import TTS
+
+CHECKPOINT = "best_model.pth"
+MODEL_DIR = Path(__file__).parent
+
+
+def clean_text(text: str) -> str:
+    text = re.sub(r"(?:\{[^}]*\}\s*)+", lambda m: re.findall(r"\{([^}]*)\}", m.group(0))[-1].strip(), text)
+    text = unicodedata.normalize("NFC", text)
+    text = text.replace(":", "").replace(".", "").replace("'", "").replace("’", "")
+    return re.sub(r"  +", " ", text).strip()
+
+
+def synthesize(tts, text: str, output_path: str) -> None:
+    tts.tts_to_file(text=text, file_path=output_path)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Male Tamil-script TTS inference")
+    parser.add_argument("text", help="Input text in Tamil script")
+    parser.add_argument("-o", "--output", default="output.wav", help="Output WAV path (default: output.wav)")
+    parser.add_argument("--gpu", type=int, default=0, help="GPU device ID (default: 0)")
+    parser.add_argument("--cpu", action="store_true", help="Force CPU inference")
+    args = parser.parse_args()
+
+    if not args.cpu:
+        os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
+
+    device = "cpu" if args.cpu else ("cuda:0" if torch.cuda.is_available() else "cpu")
+
+    tts = TTS(
+        model_path=str(MODEL_DIR / CHECKPOINT),
+        config_path=str(MODEL_DIR / "config.json"),
+        progress_bar=False,
+    )
+    tts.to(device)
+
+    text = clean_text(args.text)
+    print(f"Device    : {device}")
+    print(f"Text      : {text}")
+    print(f"Output    : {args.output}")
+
+    synthesize(tts, text, args.output)
+    print("Done.")
+
+
+if __name__ == "__main__":
+    main()

Sourashtra-Male_Script-tamil/requirements.txt

@@ -0,0 +1,4 @@
+coqui-tts
+torch
+soundfile
+numpy