Replace v2 with unchunked AvaSanj-approved 50K model

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text

README.md

@@ -19,75 +19,70 @@ pretty_name: Gooya RizehPizeh v2
 
 # Gooya RizehPizeh v2
 
-Persian (Farsi) single-speaker Piper/VITS2 voice fine-tuned on approved Gemini TTS Wikipedia audio. The model uses a case-sensitive Negara-style phonetic/Finglish frontend and produces 22,050 Hz mono audio.
+Persian single-speaker Piper/VITS2 voice trained on full, unchunked Wikipedia recordings whose audio/transcript agreement passed AvaSanj. It uses a case-sensitive Negara-style phonetic/Finglish frontend and produces 22,050 Hz mono audio.
 
-## Selected checkpoint
+## Current release
 
-This release selects the checkpoint at **global step 50,000** (`val_mel = 0.3951`). Training continued to 100,000 global steps, but later validation was worse, so the terminal state is intentionally not promoted.
+This revision replaces the earlier segmented-data v2 model with the user-selected unchunked/AvaSanj-approved 50K run.
 
+- Full checkpoint: `checkpoint/gooya-rizehpizeh-v2-unchunked-avasanj-step50000-val_mel0.4138.ckpt`
+- Checkpoint SHA-256: `8b52b3b6060b4698aba0c8bbd2f9b338aa6ba811d28a620e5f492c0fa1ae4d2b`
+- Piper/browser model: `gooya-rizehpizeh-v2.onnx`
+- ONNX SHA-256: `3f85f450fc21415e1eddaa0aa894c667e76d14aa9a342c4e77f1f9009b96f170`
+- ONNX config SHA-256: `981770210ac1aea618cb8134597127911051deabf9900824df3388f209a884fa`
 - Generator parameters: approximately 23.7M
-- Full trainable checkpoint: `checkpoint/gooya-rizehpizeh-v2-step50000-val_mel0.3951.ckpt`
-- Checkpoint SHA-256: `0c2d22718894e20fb6361c7bf2d75ee54d61db3a167983ce5d843ff6ed438938`
-- Browser/Piper model: `gooya-rizehpizeh-v2.onnx`
-- Browser/Piper model SHA-256: `675c557524a325cc014e27acb7bb9f7eab584d3247216f934e8a378b709a2d3b`
 - Sample rate: 22,050 Hz
 - Speakers: 1
 - Phoneme type: `text`
 - Symbol inventory: 256 IDs
 
-## Model lineage
+## Training data
 
-The original v1.5 production ONNX did not have a matching resumable checkpoint. A trainable hybrid checkpoint was reconstructed by copying all 350 inference-time generator tensors from the production ONNX into a compatible Mana Piper training checkpoint. Training-only branches absent from ONNX came from that Mana template.
+The parent corpus comes from the approved Persian rows in [`Reza2kn/Wikipedia-FA-EN-DeepSeek-V4-Flash-0731`](https://huggingface.co/datasets/Reza2kn/Wikipedia-FA-EN-DeepSeek-V4-Flash-0731).
 
-Before fine-tuning, a fresh ONNX round-trip matched the production model on 19/19 deterministic test rows:
+- Train: 1,290 full recordings, 21.1455 hours
+- Parent-disjoint validation: 100 full recordings, 1.6445 hours
+- Untouched sealed-eval agreement slice: 253 recordings, 4.1080 hours
+- Recording duration: 38.68–95.96 seconds; mean 58.94 seconds
+- Phoneme IDs per recording: 1,711–3,423; mean 2,174.65
+- AvaSanj compact CER threshold: `<= 0.12`
+- Train/validation/sealed-eval parent overlap: zero
+- Audio transform: none
+- Silence trimming: disabled
+- Segmentation/chunking: none
 
-- Worst maximum absolute error: `1.0348856449127197e-05`
-- Minimum SNR: `110.0768 dB`
-- Minimum correlation: `0.999999999995`
+The 253-row sealed-eval agreement slice was not consumed by training or checkpoint selection. The complete selection and hash receipt is in `training/unchunked-avasanj-agree-report.json`.
 
-The complete recovery and parity receipts are under `training/`.
+## Lineage and training
 
-## Fine-tuning data
+The original v1.5 production ONNX had no matching resumable checkpoint. Its 350 inference-time generator tensors were reconstructed into a compatible trainable Piper checkpoint; training-only branches absent from the ONNX came from the compatible Mana template. Recovery receipts remain under `training/`.
 
-The parent corpus came from approved rows in [`Reza2kn/Wikipedia-FA-EN-DeepSeek-V4-Flash-0731`](https://huggingface.co/datasets/Reza2kn/Wikipedia-FA-EN-DeepSeek-V4-Flash-0731).
+The promoted run resumed from that recovered v1.5 state and used:
 
-- Requested parent clips: 4,186 (`3,517` train, `669` sealed eval)
-- Successfully aligned parents: 4,116
-- Rejected during alignment: 70
-- Materialized segments: 30,965
-- Training segments consumed: 26,053
-- Sealed eval segments, not consumed by training: 4,912
-- Total materialized duration: 68.9853 hours
-- Segment duration: mean `8.0203s`, min `0.6307s`, max `21.6656s`
+- BF16 mixed precision
+- Batch size 1
+- Generator learning rate `5e-6`
+- Discriminator learning rate `2.5e-6`
+- Configured limit: 50,000 optimizer steps
+- Final/best validation mel: `0.4138`
 
-Audio timing came from AvaSanj v1 CTC forced alignment. Training phoneme labels came from Negara v7.1. Parent-level train/eval boundaries were preserved.
+The exact continuation script is `training/train_unchunked_avasanj_agree_full50k.sh`.
 
-## Training
+## Validation
 
-- Precision: BF16 mixed precision
-- Generator learning rate: `1e-5`
-- Discriminator learning rate: `5e-6`
-- Batch size: 1
-- Selected validation metric: `val_mel`
-- Selected checkpoint: global step 50,000, `val_mel = 0.3951`
+The published ONNX is self-contained and has no external tensor file. Direct ONNX Runtime CPU inference passed on both the canonical and long colloquial probes with finite, positive-length output. The same WAVs used for human listening are included under `samples/`.
 
-The 100K run was resumed from a successful 2K signal checkpoint. The sealed 669-parent eval split was not used for checkpoint selection.
+- Inputs: `input`, `input_lengths`, `scales`
+- Output: `output`
+- File size: 63,516,051 bytes
+- Canonical sample: 4.7137 seconds
+- Long colloquial sample: 12.6084 seconds
 
-## ONNX and browser validation
+Machine-readable receipts are under `browser/` and `training/`.
 
-The selected checkpoint was exported twice with Piper's ONNX exporter. Both exports were byte-identical. The published ONNX is self-contained and has no external tensor-data file.
+## Inference
 
-- File size: `63,516,051` bytes
-- Inputs: `input` (`int64`, `[batch, phonemes]`), `input_lengths` (`int64`, `[batch]`), `scales` (`float32`, `[3]`)
-- Output: `output` (`float32`, dynamic audio length)
-- ONNX Runtime CPU probes: 19/19 finite, positive-length outputs
-- Gouya.app `onnxruntime-web@1.27.0` WASM smoke test: passed using the application's real `rizehPhoneTextToIds` mapping
-
-Machine-readable receipts are in `browser/`.
-
-## Inference and long-form use
-
-Input must already be case-sensitive Negara-compatible phonetic text. Example:
+Input must already be case-sensitive Negara-compatible phonetic text:
 
 ```bash
 echo "zurxAneye vahdat yek makAne varzeSi dar Sahre borujerd ast." | \
@@ -96,15 +91,14 @@ echo "zurxAneye vahdat yek makAne varzeSi dar Sahre borujerd ast." | \
         --output_file output.wav
 ```
 
-For long-form narration, do not send an article as one model input. Split at semantic sentence/clause boundaries into substantial spans, synthesize each span independently, trim quiet waveform edges, and join with boundary-aware pauses. The final output duration can be arbitrarily long even though each model invocation remains bounded.
+For long-form narration, split at semantic sentence or clause boundaries into substantial spans and join them with boundary-aware pauses. Tiny fixed chunks tend to sound breathy and robotic.
 
 ## Limitations
 
-- Lower validation mel loss did not guarantee universally better pronunciation. Human listening found some words, including proper nouns, better at the earlier 2K checkpoint or the original v1.5 model.
 - G2P/Finglish quality remains the dominant source of pronunciation errors.
-- Very long single-pass phoneme sequences can collapse or omit content; use chunked synthesis.
-- Overly small chunks sound breathy and robotic. Prefer substantial semantic spans instead of fixed tiny fragments.
-- Browser export validation does not by itself constitute live Gouya.app promotion; the deployed consumer path must still be tested after rollout.
+- The approved training labels are flat Negara phoneme sequences; existing internal punctuation was not available in those labels.
+- Very long single-pass phoneme sequences can still collapse or omit content; use substantial semantic spans.
+- Proper nouns may require corrected phonetic input.
 
 ## License
 

browser/browser-wasm-smoke-report.json

@@ -1,19 +0,0 @@
-{
-  "duration_seconds": 2.9373242630385485,
-  "finite": true,
-  "input_ids": 131,
-  "model_sha256": "675c557524a325cc014e27acb7bb9f7eab584d3247216f934e8a378b709a2d3b",
-  "output_dims": [
-    1,
-    1,
-    1,
-    64768
-  ],
-  "output_samples": 64768,
-  "peak": 0.7937009930610657,
-  "provider": "wasm",
-  "runtime": "onnxruntime-web@1.27.0",
-  "schema_version": "gooya.rizehpizeh-v2-browser-smoke/v1",
-  "status": "accepted",
-  "test_phone_text": "zurxAneye vahdat yek makAne varzeSi dar Sahre borujerd ast."
-}

browser/onnx-export-report.json

@@ -1,314 +0,0 @@
-{
-  "bytes": 63516051,
-  "config_sha256": "981770210ac1aea618cb8134597127911051deabf9900824df3388f209a884fa",
-  "deterministic_scales": [
-    0.0,
-    1.0,
-    0.0
-  ],
-  "external_data": false,
-  "inputs": [
-    {
-      "name": "input",
-      "shape": [
-        "batch_size",
-        "phonemes"
-      ],
-      "type": "tensor(int64)"
-    },
-    {
-      "name": "input_lengths",
-      "shape": [
-        "batch_size"
-      ],
-      "type": "tensor(int64)"
-    },
-    {
-      "name": "scales",
-      "shape": [
-        3
-      ],
-      "type": "tensor(float)"
-    }
-  ],
-  "ir_version": 8,
-  "onnx_sha256": "675c557524a325cc014e27acb7bb9f7eab584d3247216f934e8a378b709a2d3b",
-  "opsets": [
-    {
-      "domain": "",
-      "version": 15
-    }
-  ],
-  "outputs": [
-    {
-      "name": "output",
-      "shape": [
-        "batch_size",
-        1,
-        1,
-        "Unsqueezeoutput_dim_3"
-      ],
-      "type": "tensor(float)"
-    }
-  ],
-  "providers": [
-    "CPUExecutionProvider"
-  ],
-  "rows": [
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        137728
-      ],
-      "peak": 0.8556598424911499,
-      "phone_ids": 321,
-      "samples": 137728,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--000.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        133888
-      ],
-      "peak": 0.7498155236244202,
-      "phone_ids": 319,
-      "samples": 133888,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--001.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        136448
-      ],
-      "peak": 0.8162416219711304,
-      "phone_ids": 287,
-      "samples": 136448,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--002.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        129280
-      ],
-      "peak": 0.7922796010971069,
-      "phone_ids": 325,
-      "samples": 129280,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--003.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        141312
-      ],
-      "peak": 0.8399786353111267,
-      "phone_ids": 217,
-      "samples": 141312,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--004.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        147456
-      ],
-      "peak": 0.839239776134491,
-      "phone_ids": 253,
-      "samples": 147456,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--005.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        62464
-      ],
-      "peak": 0.8430166244506836,
-      "phone_ids": 113,
-      "samples": 62464,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0014-11c7af4978b2--006.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        150016
-      ],
-      "peak": 0.8254867792129517,
-      "phone_ids": 407,
-      "samples": 150016,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--000.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        84480
-      ],
-      "peak": 0.8269263505935669,
-      "phone_ids": 197,
-      "samples": 84480,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--001.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        135680
-      ],
-      "peak": 0.7935409545898438,
-      "phone_ids": 301,
-      "samples": 135680,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--002.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        122880
-      ],
-      "peak": 0.8605506420135498,
-      "phone_ids": 235,
-      "samples": 122880,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--003.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        122624
-      ],
-      "peak": 0.7984231114387512,
-      "phone_ids": 289,
-      "samples": 122624,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--004.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        142336
-      ],
-      "peak": 0.8614963293075562,
-      "phone_ids": 321,
-      "samples": 142336,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--005.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        48384
-      ],
-      "peak": 0.8291678428649902,
-      "phone_ids": 99,
-      "samples": 48384,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0020-badd7f9a1cf9--006.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        118784
-      ],
-      "peak": 0.7483747005462646,
-      "phone_ids": 305,
-      "samples": 118784,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0021-d0f01be99058--000.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        141056
-      ],
-      "peak": 0.8492122888565063,
-      "phone_ids": 323,
-      "samples": 141056,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0021-d0f01be99058--001.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        151552
-      ],
-      "peak": 0.7924327254295349,
-      "phone_ids": 277,
-      "samples": 151552,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0021-d0f01be99058--002.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        133120
-      ],
-      "peak": 0.8153193593025208,
-      "phone_ids": 291,
-      "samples": 133120,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0021-d0f01be99058--003.flac"
-    },
-    {
-      "finite": true,
-      "output_shape": [
-        1,
-        1,
-        1,
-        118528
-      ],
-      "peak": 0.7740210890769958,
-      "phone_ids": 267,
-      "samples": 118528,
-      "utterance": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/audio/eval/wiki-fa-0021-d0f01be99058--004.flac"
-    }
-  ],
-  "schema_version": "gooya.rizehpizeh-v2-onnx-export/v1",
-  "source_checkpoint_sha256": "0c2d22718894e20fb6361c7bf2d75ee54d61db3a167983ce5d843ff6ed438938",
-  "status": "accepted",
-  "summary": {
-    "finite_rows": 19,
-    "positive_length_rows": 19,
-    "rows": 19
-  }
-}

browser/onnx-runtime-smoke-report.json

@@ -0,0 +1,22 @@
+{
+  "schema_version": "gooya.rizehpizeh-v2-unchunked-onnx-smoke/v1",
+  "model_sha256": "3f85f450fc21415e1eddaa0aa894c667e76d14aa9a342c4e77f1f9009b96f170",
+  "config_sha256": "981770210ac1aea618cb8134597127911051deabf9900824df3388f209a884fa",
+  "bytes": 63516051,
+  "onnxruntime_version": "1.23.2",
+  "providers": ["CPUExecutionProvider"],
+  "inputs": [
+    {"name": "input", "type": "tensor(int64)", "shape": ["batch_size", "phonemes"]},
+    {"name": "input_lengths", "type": "tensor(int64)", "shape": ["batch_size"]},
+    {"name": "scales", "type": "tensor(float)", "shape": [3]}
+  ],
+  "outputs": [
+    {"name": "output", "type": "tensor(float)", "shape": ["batch_size", 1, 1, "Unsqueezeoutput_dim_3"]}
+  ],
+  "scales": [0.667, 1, 0.8],
+  "rows": [
+    {"name": "canonical", "phone_ids": 181, "output_shape": [1, 1, 1, 99328], "samples": 99328, "finite": true, "peak": 0.8139110207557678},
+    {"name": "long_colloquial", "phone_ids": 571, "output_shape": [1, 1, 1, 283136], "samples": 283136, "finite": true, "peak": 0.8082336783409119}
+  ],
+  "passed": true
+}

checkpoint/{gooya-rizehpizeh-v2-step50000-val_mel0.3951.ckpt → gooya-rizehpizeh-v2-unchunked-avasanj-step50000-val_mel0.4138.ckpt}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c2d22718894e20fb6361c7bf2d75ee54d61db3a167983ce5d843ff6ed438938
-size 845890322
+oid sha256:8b52b3b6060b4698aba0c8bbd2f9b338aa6ba811d28a620e5f492c0fa1ae4d2b
+size 845890386

gooya-rizehpizeh-v2.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:675c557524a325cc014e27acb7bb9f7eab584d3247216f934e8a378b709a2d3b
+oid sha256:3f85f450fc21415e1eddaa0aa894c667e76d14aa9a342c4e77f1f9009b96f170
 size 63516051

release-manifest.json

@@ -1,73 +1,48 @@
 {
-  "files": [
-    {
-      "bytes": 1519,
-      "path": ".gitattributes",
-      "sha256": "11ad7efa24975ee4b0c3c3a38ed18737f0658a5f75a0a96787b576a78a023361"
-    },
-    {
-      "bytes": 1064,
-      "path": "LICENSE",
-      "sha256": "beacc839712cc8db59ad2997b52edb77946461593219d82276e39ba7e404f7ef"
-    },
-    {
-      "bytes": 4933,
-      "path": "README.md",
-      "sha256": "8699cb2c567fb8c0c4a52ea9e0edb691bbd79c7f44d81b1b52f75bdaf10e3439"
-    },
-    {
-      "bytes": 510,
-      "path": "browser/browser-wasm-smoke-report.json",
-      "sha256": "ce1202b0675cf3c72944334bb1fbd61997af34e7ddc2d15166fd35b617cd54a0"
-    },
-    {
-      "bytes": 7588,
-      "path": "browser/onnx-export-report.json",
-      "sha256": "e1812cdfd6dd31768245b90f13e47b6785346e9f1b09083cb6bfd57b7d093d0f"
-    },
-    {
-      "bytes": 845890322,
-      "path": "checkpoint/gooya-rizehpizeh-v2-step50000-val_mel0.3951.ckpt",
-      "sha256": "0c2d22718894e20fb6361c7bf2d75ee54d61db3a167983ce5d843ff6ed438938"
-    },
-    {
-      "bytes": 3642,
-      "path": "config/negara-phoneme-id-map.json",
-      "sha256": "6e67148a46b9bf5a3529c97cc1fa60a10f477d0bd80e61a1fb1a24ad59eefe83"
-    },
-    {
+  "schema_version": "gooya.rizehpizeh-v2-release/v2",
+  "release": "unchunked-avasanj-agree-50k",
+  "model_repo": "Reza2kn/Gooya-RizehPizeh-v2",
+  "sample_rate": 22050,
+  "files": {
+    "checkpoint/gooya-rizehpizeh-v2-unchunked-avasanj-step50000-val_mel0.4138.ckpt": {
+      "bytes": 845890386,
+      "sha256": "8b52b3b6060b4698aba0c8bbd2f9b338aa6ba811d28a620e5f492c0fa1ae4d2b"
+    },
+    "gooya-rizehpizeh-v2.onnx": {
       "bytes": 63516051,
-      "path": "gooya-rizehpizeh-v2.onnx",
-      "sha256": "675c557524a325cc014e27acb7bb9f7eab584d3247216f934e8a378b709a2d3b"
+      "sha256": "3f85f450fc21415e1eddaa0aa894c667e76d14aa9a342c4e77f1f9009b96f170"
     },
-    {
+    "gooya-rizehpizeh-v2.onnx.json": {
       "bytes": 5000,
-      "path": "gooya-rizehpizeh-v2.onnx.json",
       "sha256": "981770210ac1aea618cb8134597127911051deabf9900824df3388f209a884fa"
     },
-    {
-      "bytes": 1903,
-      "path": "training/alignment-report.json",
-      "sha256": "b8dff1be540f03404b05d552eac009f5ca1781c5190898622851331ac44de325"
+    "samples/canonical-intro.wav": {
+      "bytes": 207916,
+      "sha256": "d9978de38d5d9db03aa2dcedbaa357bd88c4fd2ec8ede8b3ac8c088e50682713"
     },
-    {
-      "bytes": 9626,
-      "path": "training/base-onnx-roundtrip-parity.json",
-      "sha256": "1e1346518a1cd72c7b8d9368845c89206ba026d78417af753a32f9c41fdaef59"
+    "samples/long-colloquial.wav": {
+      "bytes": 556076,
+      "sha256": "cc0c864c5b4d414f7b35d9506dacc120d2fafc15fd9c29f9be55c60e5525aa6d"
+    },
+    "config/negara-phoneme-id-map.json": {
+      "sha256": "6e67148a46b9bf5a3529c97cc1fa60a10f477d0bd80e61a1fb1a24ad59eefe83"
     },
-    {
-      "bytes": 82458,
-      "path": "training/checkpoint-recovery-report.json",
-      "sha256": "ec0bb47bbf07c5bfeff92ab68ec57e1099d94cd5172d087f8f6694e5ce20e178"
+    "training/unchunked-avasanj-agree-report.json": {
+      "sha256": "e03203d01d94421a08e65c64806f40d084085714d8e260982f0e8e7a888c3ee6"
     },
-    {
-      "bytes": 2913,
-      "path": "training/train_rizeh_full_100k.sh",
-      "sha256": "3b7a08aee6af72a787000a62b209fdf70c8726a65a8a5e66176c4106158f392c"
+    "training/train_unchunked_avasanj_agree_full50k.sh": {
+      "sha256": "d4c08c3cec513bd6cb829b91ed111bc8656283d47a482bd8fe850c590f7c571a"
     }
-  ],
-  "schema_version": "gooya.rizehpizeh-v2-release/v1",
-  "selected_global_step": 50000,
-  "selected_val_mel": 0.3951,
-  "status": "checkpoint-and-browser-onnx-ready"
+  },
+  "training": {
+    "train_parents": 1290,
+    "validation_parents": 100,
+    "sealed_eval_agree_parents": 253,
+    "parent_overlap": 0,
+    "audio_chunking": false,
+    "trim_silence": false,
+    "avasanj_compact_cer_max_inclusive": 0.12,
+    "max_optimizer_steps": 50000,
+    "best_val_mel": 0.4137682318687439
+  }
 }

samples/canonical-intro.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9978de38d5d9db03aa2dcedbaa357bd88c4fd2ec8ede8b3ac8c088e50682713
+size 207916

samples/long-colloquial.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc0c864c5b4d414f7b35d9506dacc120d2fafc15fd9c29f9be55c60e5525aa6d
+size 556076

training/alignment-report.json

@@ -1,48 +0,0 @@
-{
-  "models": {
-    "avasanj_model_sha256": "c7ea4f6c1e510e4c23bbc1856c290787f20a730d14c4d95de45c03259d5dd07c",
-    "mana_config_sha256": "df13724ca155a8699d8c42d47d5748dee98ec8057cafdaed8b1455fc85798753",
-    "mana_phoneme_map_sha256": "ba38263aec458bca093a0e60629a442dfb365bd7ef79e70be58c311ba886f868",
-    "negara_model_sha256": "2d89883abe07082cfb291203c1e5609f8974e71ca04d6c3e73bf2d32f8b715cf",
-    "negara_overlay_sha256": "9e063c2f26db4f56f6a93fef33f1f2a25423318ff7073b9954fddf4c4085412f",
-    "negara_runtime_map_sha256": "6e67148a46b9bf5a3529c97cc1fa60a10f477d0bd80e61a1fb1a24ad59eefe83"
-  },
-  "outputs": {
-    "eval_metadata_sha256": "587ff471a78e61f471d524435f9bbe9c750378861185bf27b88656d27d41652d",
-    "segments_sha256": "f7fd2abb87bd73b4e80dc8a5cd23acb9a8fafd480d56ab9d500bacbbb7b5d4ed",
-    "train_metadata_sha256": "3077e23e86a97454b36c1d5c7d2b34108b332752b3f600811418714af8928320"
-  },
-  "parent_manifest": "/home/rezo/gooya-bozorg-v15-gemini-approved-20260829/training_manifest.jsonl",
-  "parent_manifest_sha256": "d285458025385945d10e733bdea74080f544c4aa3c621454686d6ffe7678f345",
-  "parent_rows_requested": 4186,
-  "parent_split_counts": {
-    "eval": 669,
-    "train": 3517
-  },
-  "parent_status_counts": {
-    "complete": 4116,
-    "rejected": 70
-  },
-  "schema_version": "gooya.piper-wikipedia-aligned/v1",
-  "segment_duration_seconds": {
-    "max": 21.665583333333334,
-    "mean": 8.020255083696643,
-    "min": 0.6307083333333333
-  },
-  "segment_hours": 68.98533296296293,
-  "segment_rows": 30965,
-  "segment_split_counts": {
-    "eval": 4912,
-    "train": 26053
-  },
-  "segmentation": {
-    "audio_timing_source": "AvaSanj v1 CTC forced alignment",
-    "edge_padding_seconds": 0.08,
-    "maximum_seconds": 12.0,
-    "minimum_seconds": 2.0,
-    "parent_split_preserved": true,
-    "target_seconds": 8.0,
-    "training_label_source": "Negara v7.1"
-  },
-  "status": "complete"
-}

training/train_rizeh_full_100k.sh

@@ -1,82 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-root=/home/rezo/piper-wikipedia-approved-20260830
-data="${root}/data/aligned-full-v1"
-signal_run="${root}/runs/rizeh-recovered-approved-signal-2000"
-run="${root}/runs/rizeh-recovered-approved-full-100k"
-resume_checkpoint="${signal_run}/lightning_logs/version_1/checkpoints/last.ckpt"
-
-[[ -s "${resume_checkpoint}" ]]
-[[ "$(jq -r .status "${data}/report.json")" == complete ]]
-mkdir -p "${run}"
-
-cd "${root}/src/piper1-gpl"
-"${root}/venv/bin/python" -m piper.train fit \
-  --trainer.accelerator gpu \
-  --trainer.devices 1 \
-  --trainer.precision bf16-mixed \
-  --trainer.max_steps 100000 \
-  --trainer.num_sanity_val_steps 0 \
-  --trainer.val_check_interval 500 \
-  --trainer.limit_val_batches 25 \
-  --trainer.log_every_n_steps 10 \
-  --trainer.enable_progress_bar false \
-  --trainer.default_root_dir "${run}" \
-  --model.sample_rate 22050 \
-  --model.num_speakers 1 \
-  --model.warmstart_ckpt null \
-  --model.learning_rate 0.00001 \
-  --model.learning_rate_d 0.000005 \
-  --model.mos_metric null \
-  --data.voice_name gooya-rizehpizeh-wikipedia-kore-full-100k \
-  --data.csv_path "${data}/metadata-train.csv" \
-  --data.audio_dir / \
-  --data.cache_dir "${root}/cache/aligned-full-v1" \
-  --data.config_path "${signal_run}/model.onnx.json" \
-  --data.batch_size 1 \
-  --data.validation_split 0.01 \
-  --data.num_test_examples 0 \
-  --data.num_workers 2 \
-  --data.trim_silence false \
-  --data.phoneme_type text \
-  --data.dataset_type phoneme_ids \
-  --data.phonemes_path "${data}/negara-phoneme-id-map.json" \
-  --data.espeak_voice fa \
-  --data.num_symbols 256 \
-  --ckpt_path "${resume_checkpoint}"
-
-last_checkpoint=$(find "${run}" -name last.ckpt -type f -printf '%T@ %p\n' | sort -nr | head -1 | cut -d' ' -f2-)
-best_checkpoint=$("${root}/venv/bin/python" - "${run}" <<'PY'
-import re
-import sys
-from pathlib import Path
-
-run = Path(sys.argv[1])
-candidates = []
-for path in run.glob("lightning_logs/version_*/checkpoints/*val_mel=*.ckpt"):
-    match = re.search(r"val_mel=([0-9.]+)\.ckpt$", path.name)
-    if match:
-        candidates.append((float(match.group(1)), str(path)))
-if not candidates:
-    raise SystemExit("No validation checkpoint found")
-print(min(candidates)[1])
-PY
-)
-[[ -s "${last_checkpoint}" ]]
-[[ -s "${best_checkpoint}" ]]
-
-"${root}/venv/bin/python" -m piper.train.export_onnx \
-  --checkpoint "${best_checkpoint}" \
-  --output-file "${run}/model-best.onnx"
-
-printf '%s\n' 'zurxAneye vahdat yek makAne varzeSi dar Sahre borujerd ast.' | \
-  "${root}/venv/bin/python" -m piper \
-  -m "${run}/model-best.onnx" \
-  -f "${run}/zurxane-full-100k-best.wav" \
-  --length-scale 1.15
-
-sha256sum "${last_checkpoint}" "${best_checkpoint}" "${run}/model-best.onnx" \
-  "${run}/model-best.onnx.json" "${run}/zurxane-full-100k-best.wav" \
-  > "${run}/sha256sums.txt"
-echo "full_complete_utc=$(date -u +%FT%TZ) last=${last_checkpoint} best=${best_checkpoint}"

training/train_unchunked_avasanj_agree_full50k.sh

@@ -0,0 +1,81 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+root=/home/rezo/piper-wikipedia-approved-20260830
+data="${root}/data/unchunked-avasanj-agree-v1"
+run="${root}/runs/rizeh-unchunked-avasanj-agree-gate500"
+cache="${root}/cache/unchunked-avasanj-agree-v1"
+resume="${run}/lightning_logs/version_0/checkpoints/last.ckpt"
+
+exec 9>"${root}/runs/rizeh-unchunked-avasanj-agree-full50k.lock"
+flock -n 9 || { echo "full50k is already running" >&2; exit 73; }
+
+[[ "$(jq -r .status "${data}/report.json")" == complete ]]
+[[ "$(jq -r .rows.train "${data}/report.json")" == 1290 ]]
+[[ "$(jq -r .rows.validation "${data}/report.json")" == 100 ]]
+[[ "$(jq -r '.parent_overlap.train.validation' "${data}/report.json")" == 0 ]]
+[[ "$(wc -l < "${data}/metadata-fit-parent-disjoint.csv")" == 1390 ]]
+[[ -s "${resume}" ]]
+[[ -s "${run}/model.onnx.json" ]]
+
+cd "${root}/src/piper1-gpl"
+export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
+"${root}/venv/bin/python" -m piper.train fit \
+  --ckpt_path "${resume}" \
+  --trainer.accelerator gpu \
+  --trainer.devices 1 \
+  --trainer.precision bf16-mixed \
+  --trainer.max_steps 50000 \
+  --trainer.num_sanity_val_steps 0 \
+  --trainer.val_check_interval 500 \
+  --trainer.limit_val_batches 25 \
+  --trainer.log_every_n_steps 10 \
+  --trainer.enable_progress_bar false \
+  --trainer.default_root_dir "${run}" \
+  --model.sample_rate 22050 \
+  --model.num_speakers 1 \
+  --model.warmstart_ckpt "${root}/models/rizeh-onnx-recovery/rizehpizeh-epoch4-recovered-trainable.ckpt" \
+  --model.learning_rate 0.000005 \
+  --model.learning_rate_d 0.0000025 \
+  --model.mos_metric null \
+  --data.voice_name gooya-rizehpizeh-unchunked-avasanj-agree-full50k \
+  --data.csv_path "${data}/metadata-fit-parent-disjoint.csv" \
+  --data.audio_dir / \
+  --data.cache_dir "${cache}" \
+  --data.config_path "${run}/model.onnx.json" \
+  --data.batch_size 1 \
+  --data.validation_split 0.07194244604316546 \
+  --data.num_test_examples 0 \
+  --data.num_workers 2 \
+  --data.trim_silence false \
+  --data.phoneme_type text \
+  --data.dataset_type phoneme_ids \
+  --data.phonemes_path "${root}/data/aligned-full-v1/negara-phoneme-id-map.json" \
+  --data.espeak_voice fa \
+  --data.num_symbols 256
+
+checkpoint=$(find "${run}" -name last.ckpt -type f -printf '%T@ %p\n' | sort -nr | head -1 | cut -d' ' -f2-)
+[[ -s "${checkpoint}" ]]
+"${root}/venv/bin/python" -m piper.train.export_onnx \
+  --checkpoint "${checkpoint}" \
+  --output-file "${run}/model-step50000.onnx"
+cp "${run}/model.onnx.json" "${run}/model-step50000.onnx.json"
+
+printf '%s\n' 'salAm! man guyA hastam va in sedA haminjA dAxele morurare SomA sAxte Sode ast.' | \
+  "${root}/venv/bin/python" -m piper \
+  -m "${run}/model-step50000.onnx" \
+  -f "${run}/canonical-intro-step50000.wav" \
+  --length-scale 1.0
+
+printf '%s\n' 'salAm az beyne dostAn kasi sAyte rAyegAne tabdile matn be goftAr besurate xodemuni va mohAverei be zabAne fArsi bege miSnAse moQarrefi kone takid mikonam xorujiye fAyl be surate moddat zamAne tulAni mixAham mahdudiyyate moddate zamAn nadASte bASe.' | \
+  "${root}/venv/bin/python" -m piper \
+  -m "${run}/model-step50000.onnx" \
+  -f "${run}/long-colloquial-step50000.wav" \
+  --length-scale 1.0
+
+sha256sum "${checkpoint}" "${run}/model-step50000.onnx" \
+  "${run}/model-step50000.onnx.json" \
+  "${run}/canonical-intro-step50000.wav" \
+  "${run}/long-colloquial-step50000.wav" \
+  > "${run}/sha256sums-step50000.txt"
+echo "full50k_complete_utc=$(date -u +%FT%TZ) checkpoint=${checkpoint}"

training/unchunked-avasanj-agree-report.json

@@ -0,0 +1,68 @@
+{
+  "duration_seconds": {
+    "max": 95.96,
+    "mean": 58.93640900791231,
+    "min": 38.68
+  },
+  "excluded_reasons": {
+    "avasanj_not_agree": 2473,
+    "parent_alignment_not_complete": 70
+  },
+  "hours": {
+    "sealed_eval_agree": 4.107966666666665,
+    "train": 21.145477777777785,
+    "validation": 1.6444777777777777
+  },
+  "inputs": {
+    "mana_phoneme_map": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/mana-phoneme-id-map.json",
+    "mana_phoneme_map_sha256": "ba38263aec458bca093a0e60629a442dfb365bd7ef79e70be58c311ba886f868",
+    "manifest": "/home/rezo/gooya-bozorg-v15-gemini-approved-20260829/training_manifest.jsonl",
+    "manifest_sha256": "d285458025385945d10e733bdea74080f544c4aa3c621454686d6ffe7678f345",
+    "parents": "/home/rezo/piper-wikipedia-approved-20260830/data/aligned-full-v1/parents.jsonl",
+    "parents_sha256": "09f49adf75bc6000b82a81c956555ec00c5a1f2d7a77f6d1ea3d989a2658129a"
+  },
+  "original_manifest_splits": {
+    "eval": 669,
+    "train": 3517
+  },
+  "outputs": {
+    "excluded-parents.jsonl": "e508a4ecc721d3b7604c118cc2bdd32990e1c9e0ab93facb4af8d307136f43eb",
+    "metadata-fit-parent-disjoint.csv": "39b38a8de0ec8296d60a94c4060e151918e84d91484fa2a0f454a0c1e6828dbd",
+    "metadata-sealed_eval_agree.csv": "6bfa808cba6208939bb8fbe61429a8af7a2baab0471ec632bffab1c392527358",
+    "metadata-train.csv": "84376168ab6f6aa989604776a2d59b4d8b68a07fdd328d2e0545c556bf855437",
+    "metadata-validation.csv": "9c0f873fdd78f67a06c000c2bf4b8543f0ed32dc67e95e72ae7e7c95afaa0517",
+    "selected-parents.jsonl": "65e4559957bcdcb301238a04a09681a7070083d854e7546e80fc7949b7ab12b7"
+  },
+  "parent_overlap": {
+    "sealed_eval_agree": {
+      "train": 0,
+      "validation": 0
+    },
+    "train": {
+      "sealed_eval_agree": 0,
+      "validation": 0
+    },
+    "validation": {
+      "sealed_eval_agree": 0,
+      "train": 0
+    }
+  },
+  "phoneme_id_count": {
+    "max": 3423,
+    "mean": 2174.6494217894096,
+    "min": 1711
+  },
+  "rows": {
+    "sealed_eval_agree": 253,
+    "train": 1290,
+    "validation": 100
+  },
+  "schema_version": "gooya.piper-unchunked-avasanj-agree/v1",
+  "selection": {
+    "audio_transform": "none",
+    "internal_punctuation": "not_present_in_existing_flat_Negara_v7.1_labels",
+    "max_avasanj_compact_cer_inclusive": 0.12,
+    "terminal_punctuation": "preserved_from_source_when_present_else_period"
+  },
+  "status": "complete"
+}