VoiceHub/dacvae-tts-tr-nano-a

DACVAE-TTS Turkish run A (nano recipe, full data)

Generated audio of every evaluated checkpoint of the training run tr-nano-a (Turkish zero-shot voice-cloning TTS, dacvae-tts, frozen Meta DACVAE latents, 48 kHz). This repository holds model outputs and metrics, not training data. Training data: Vyvo/tr-dataset-12 (Turkish podcast segments).

Run A: configs/nano_tr.yaml (51.4M parameters), 17 shards of Vyvo/tr-dataset-12 with quality >= 55 (~77 h train), one RTX 4090, frame budget 12000, 40k updates. Includes sampler sweeps (guidance 2-6, 16/32 steps, sway, noise scale, duration scale) on the 15k and 20k checkpoints.

How to listen

• monitor/prompt-<uid>.wav: the reference voice given to the model (a real validation recording of an unseen speaker, decoded through the DACVAE codec, so it also shows the codec's own quality ceiling).
• monitor/<step-folder>/NNN.wav: the model's synthesis of case NNN — the text of another recording of the same speaker, in the prompt's voice. monitor/<step-folder>/results.jsonl lists per case: text (target transcript), prompt_uid (which prompt WAV), hypothesis (Whisper-large-v3 transcript of the synthesis), wer, cer, speaker_similarity, duration_ratio. NNN.json holds the sampler settings and timings.
• Folder name suffix -gG-uU-nN-sS-dD-kK: guidance G, guidance applied while t < U, initial-noise scale N, sway S, duration scale D, Euler steps K (no suffix = guidance 2, 16 steps).
• monitor/cases.json: the 48 (prompt, target) cases; monitor.jsonl: one summary row per evaluated checkpoint/setting; train.jsonl: training/validation curves; config.json: the full training configuration.

WER/CER: faster-whisper large-v3 (Turkish), Turkish text normalization (numbers spelled out, İ/ı-aware lower-casing, punctuation removed). SIM: microsoft/wavlm-base-plus-sv cosine between synthesis and the codec-decoded prompt. Corpus WER/CER over 48 cases of 10 held-out speakers; the ASR floor on real codec-decoded speech is WER ≈ 0.05 / CER ≈ 0.016.

Results

step	folder	WER	CER	SIM	duration ratio	guidance	steps
5000	monitor/step-0005000-g3-u1-n1-s-1-d1-k32	0.861	0.571	0.955	1.03	3.0	32
10000	monitor/step-0010000-g3-u1-n1-s-1-d1-k32	0.570	0.347	0.964	1.03	3.0	32
15000	monitor/step-0015000-g2-u1-n1-s-1-d1-k32	0.466	0.278	0.963	1.03	2.0	32
15000	monitor/step-0015000-g3-u1-n1-s-1-d1-k32	0.399	0.235	0.963	1.03	3.0	32
15000	monitor/step-0015000-g3-u0.5-n1-s-1-d1-k32	0.399	0.235	0.963	1.03	3.0	32
15000	monitor/step-0015000-g3-u1-n0.9-s-1-d1-k32	0.417	0.249	0.964	1.03	3.0	32
15000	monitor/step-0015000-g3-u1-n1-s0-d1-k32	0.408	0.236	0.965	1.03	3.0	32
15000	monitor/step-0015000-g3-u1-n1-s-1-d0.9-k32	0.425	0.244	0.964	0.93	3.0	32
15000	monitor/step-0015000-g3-u1-n1-s-1-d1-k16	0.394	0.233	0.964	1.03	3.0	16
15000	monitor/step-0015000-g4-u1-n1-s-1-d1-k32	0.368	0.211	0.965	1.03	4.0	32
20000	monitor/step-0020000-g3-u1-n1-s-1-d1-k32	0.305	0.184	0.966	1.03	3.0	32
20000	monitor/step-0020000-g4-u1-n1-s-1-d1-k32	0.274	0.163	0.965	1.03	4.0	32
20000	monitor/step-0020000-g4-u0.5-n1-s-1-d1-k16	0.302	0.172	0.963	1.03	4.0	16
20000	monitor/step-0020000-g5-u1-n1-s-1-d1-k32	0.257	0.157	0.963	1.03	5.0	32
20000	monitor/step-0020000-g6-u1-n1-s-1-d1-k32	0.238	0.142	0.962	1.03	6.0	32
25000	monitor/step-0025000-g3-u1-n1-s-1-d1-k32	0.256	0.148	0.965	1.03	3.0	32
30000	monitor/step-0030000-g3-u1-n1-s-1-d1-k32	0.223	0.126	0.962	1.03	3.0	32
35000	monitor/step-0035000-g3-u1-n1-s-1-d1-k32	0.181	0.106	0.948	1.03	3.0	32
40000	monitor/step-0040000-g3-u1-n1-s-1-d1-k32	0.205	0.113	0.947	1.03	3.0	32
40000	monitor/step-0040000-g3-u1-n1-s-1-d1-k32	0.202	0.113	0.947	1.03	3.0	32
40000	monitor/step-0040000-g4-u1-n1-s-1-d1-k32	0.180	0.110	0.946	1.03	4.0	32
40000	monitor/step-0040000-g5-u1-n1-s-1-d1-k32	0.162	0.099	0.946	1.03	5.0	32
40000	monitor/step-0040000-g6-u1-n1-s-1-d1-k32	0.148	0.092	0.945	1.03	6.0	32

Configuration

{
  "latent_dim": 128,
  "width": 448,
  "depth": 12,
  "heads": 7,
  "text_depth": 4,
  "patch_size": 1,
  "ff_mult": 3,
  "cond_dropout": 0.2,
  "reference_encoder": "mlp",
  "reference_pooling": "mean",
  "reference_paths": "both",
  "duration_features": "baseline",
  "positions": "rope",
  "qk_norm": true,
  "text_attention": 4,
  "prediction": "edm",
  "text_layout": "joined",
  "duration": "rule",
  "ctc_layer": 8,
  "adaln_rank": 64
}

{
  "steps": 40000,
  "batch_size": 96,
  "accumulation": 1,
  "learning_rate": 0.0008,
  "warmup": 2000,
  "weight_decay": 0.01,
  "optimizer": "muon",
  "muon_momentum": 0.95,
  "ema_decay": 0.9999,
  "precision": "bf16",
  "workers": 4,
  "worker_threads": 1,
  "prefetch_factor": 4,
  "loader_start_method": "spawn",
  "cuda_prefetch": true,
  "checkpoint_every": 1000,
  "validate_every": 1000,
  "log_every": 100,
  "grad_checkpoint": true,
  "compile": false,
  "seed": 42,
  "flow_reduction": "frame",
  "duration_weight": 0.1,
  "speaker_balance": 0.0,
  "diagnostics_every": 0,
  "pairing": "within",
  "prompt_fraction_min": 0.1,
  "prompt_fraction_max": 0.6,
  "prompt_dropout": 0.3,
  "time_sampling": "logit_normal",
  "batch_expansion": 2,
  "keep_every": 5000,
  "ctc_weight": 0.1,
  "contrastive_weight": 0.2,
  "contrastive_margin": 0.1,
  "wandb_project": "dacvae-tts-tr"
}