--- dataset_info: - config_name: aishell3 features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1434674764.444 num_examples: 2814 download_size: 1298956633 dataset_size: 1434674764.444 - config_name: argentinian_es features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2401506764.784 num_examples: 4362 download_size: 1953967499 dataset_size: 2401506764.784 - config_name: basque features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 4510039899.312 num_examples: 6907 download_size: 3694399344 dataset_size: 4510039899.312 - config_name: burmese features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1263153045.248 num_examples: 2071 download_size: 1030489997 dataset_size: 1263153045.248 - config_name: catalan features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 3304304924.488 num_examples: 4018 download_size: 2396720666 dataset_size: 3304304924.488 - config_name: chilean_es features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2344832209.28 num_examples: 3755 download_size: 1733908148 dataset_size: 2344832209.28 - config_name: colombian_es features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2396032660.56 num_examples: 4105 download_size: 1853162463 dataset_size: 2396032660.56 - config_name: cv_ta features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 126859040.0 num_examples: 263 download_size: 112407214 dataset_size: 126859040.0 - config_name: cv_zhcn features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 6015322537.848 num_examples: 8948 download_size: 4738445424 dataset_size: 6015322537.848 - config_name: daps features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 4433979121.233 num_examples: 3079 download_size: 4219387493 dataset_size: 4433979121.233 - config_name: galician features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 3476292735.872 num_examples: 5343 download_size: 2690014565 dataset_size: 3476292735.872 - config_name: gujarati features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2692727345.904 num_examples: 3789 download_size: 2048986049 dataset_size: 2692727345.904 - config_name: hifitts features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 10105661759.938 num_examples: 19447 download_size: 10167653025 dataset_size: 10105661759.938 - config_name: hui_german features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 24422866511.75 num_examples: 28359 download_size: 23623627005 dataset_size: 24422866511.75 - config_name: javanese features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1455115751.828 num_examples: 2906 download_size: 1368757011 dataset_size: 1455115751.828 - config_name: kannada features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2636611262.92 num_examples: 3542 download_size: 2110839026 dataset_size: 2636611262.92 - config_name: khmer features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1109767557.416 num_examples: 2098 download_size: 856044100 dataset_size: 1109767557.416 - config_name: kss_korean features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1826827766.564 num_examples: 3994 download_size: 1648909927 dataset_size: 1826827766.564 - config_name: malayalam features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1492333104.46 num_examples: 2470 download_size: 1197392249 dataset_size: 1492333104.46 - config_name: marathi features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 972771196.6 num_examples: 1400 download_size: 821207252 dataset_size: 972771196.6 - config_name: nepali features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 693370714.84 num_examples: 1157 download_size: 669613252 dataset_size: 693370714.84 - config_name: nigerian_en features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1774719446.76 num_examples: 2687 download_size: 1357170267 dataset_size: 1774719446.76 - config_name: peruvian_es features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2985297200.36 num_examples: 4991 download_size: 2382720758 dataset_size: 2985297200.36 - config_name: puertorico_es features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 312198021.0 num_examples: 516 download_size: 244476478 dataset_size: 312198021.0 - config_name: ravdess features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 98138536.0 num_examples: 241 download_size: 64632225 dataset_size: 98138536.0 - config_name: south_african features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2609797704.28 num_examples: 5240 download_size: 2595787393 dataset_size: 2609797704.28 - config_name: sundanese features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1222348478.408 num_examples: 2116 download_size: 1182371107 dataset_size: 1222348478.408 - config_name: tamil features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 2328867763.712 num_examples: 3633 download_size: 1620299433 dataset_size: 2328867763.712 - config_name: telugu features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1280168624.328 num_examples: 2517 download_size: 888160549 dataset_size: 1280168624.328 - config_name: uk_ireland_en features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 9282270221.4 num_examples: 14925 download_size: 8067467815 dataset_size: 9282270221.4 - config_name: venezuelan_es features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 1452261660.888 num_examples: 2466 download_size: 1080960605 dataset_size: 1452261660.888 - config_name: yoruba features: - name: audio dtype: audio: sampling_rate: 48000 - name: source dtype: string - name: bak dtype: float32 - name: sig dtype: float32 - name: ovrl dtype: float32 - name: duration dtype: float32 splits: - name: train num_bytes: 752974618.2 num_examples: 1464 download_size: 611970946 dataset_size: 752974618.2 - config_name: labeled44k features: - name: audio dtype: audio - name: source dtype: string - name: duration dtype: float32 - name: sr dtype: int32 - name: lang dtype: string - name: speaker dtype: string - name: text dtype: string splits: - name: train num_examples: 2870194 download_size: 150319828130 configs: - config_name: aishell3 data_files: - split: train path: aishell3/train-* - config_name: argentinian_es data_files: - split: train path: argentinian_es/train-* - config_name: basque data_files: - split: train path: basque/train-* - config_name: burmese data_files: - split: train path: burmese/train-* - config_name: catalan data_files: - split: train path: catalan/train-* - config_name: chilean_es data_files: - split: train path: chilean_es/train-* - config_name: colombian_es data_files: - split: train path: colombian_es/train-* - config_name: cv_ta data_files: - split: train path: cv_ta/train-* - config_name: cv_zhcn data_files: - split: train path: cv_zhcn/train-* - config_name: daps data_files: - split: train path: daps/train-* - config_name: galician data_files: - split: train path: galician/train-* - config_name: gujarati data_files: - split: train path: gujarati/train-* - config_name: hifitts data_files: - split: train path: hifitts/train-* - config_name: hui_german data_files: - split: train path: hui_german/train-* - config_name: javanese data_files: - split: train path: javanese/train-* - config_name: kannada data_files: - split: train path: kannada/train-* - config_name: khmer data_files: - split: train path: khmer/train-* - config_name: kss_korean data_files: - split: train path: kss_korean/train-* - config_name: malayalam data_files: - split: train path: malayalam/train-* - config_name: marathi data_files: - split: train path: marathi/train-* - config_name: nepali data_files: - split: train path: nepali/train-* - config_name: nigerian_en data_files: - split: train path: nigerian_en/train-* - config_name: peruvian_es data_files: - split: train path: peruvian_es/train-* - config_name: puertorico_es data_files: - split: train path: puertorico_es/train-* - config_name: ravdess data_files: - split: train path: ravdess/train-* - config_name: south_african data_files: - split: train path: south_african/train-* - config_name: sundanese data_files: - split: train path: sundanese/train-* - config_name: tamil data_files: - split: train path: tamil/train-* - config_name: telugu data_files: - split: train path: telugu/train-* - config_name: uk_ireland_en data_files: - split: train path: uk_ireland_en/train-* - config_name: venezuelan_es data_files: - split: train path: venezuelan_es/train-* - config_name: yoruba data_files: - split: train path: yoruba/train-* - config_name: labeled44k data_files: - split: train path: labeled44k/train-* --- # TTS-Clean44k A **multilingual pool of verified-clean, wideband speech** for training and evaluating speech restoration / text-to-speech (TTS) models. Every utterance is independently checked on **two** axes and stored as parquet with its per-utterance quality scores attached: 1. **Native sample rate ≥ 44.1 kHz** — measured per file with `ffprobe`, *never* trusting the source's advertised rate. Anything below 44.1 kHz is dropped. 2. **DNSMOS P.835 `bak` ≥ 3.644** — the background-noise MOS from the [DNSMOS P.835](https://github.com/microsoft/DNS-Challenge) model. Only genuinely clean recordings pass. The dataset was assembled as the **clean teacher pool** for Sidon call-centre speech restoration (the decoder is trained to *reproduce* its teacher, so the teacher must be genuinely clean and full-band), but it is broadly useful as a filtered multilingual TTS corpus. ## Composition **28 language/source configs · 119,950 utterances · 208.7 h · utterance-weighted DNSMOS bak 4.036 / sig 3.484 / ovrl 3.203.** | # | config | utts | hours | bak | sig | ovrl | |--:|--------|-----:|------:|----:|----:|-----:| | 1 | hifitts | 19,447 | 30.0 | 4.026 | 3.533 | 3.247 | | 2 | uk_ireland_en | 14,925 | 28.0 | 4.077 | 3.572 | 3.313 | | 3 | cv_zhcn | 8,948 | 15.4 | 3.952 | 3.466 | 3.145 | | 4 | basque | 6,907 | 13.5 | 4.043 | 3.521 | 3.230 | | 5 | galician | 5,343 | 10.0 | 4.055 | 3.521 | 3.248 | | 6 | catalan | 4,018 | 9.0 | 3.977 | 3.433 | 3.127 | | 7 | peruvian_es | 4,991 | 8.8 | 4.052 | 3.423 | 3.154 | | 8 | south_african | 5,240 | 8.3 | 3.977 | 3.460 | 3.149 | | 9 | kannada | 3,542 | 7.6 | 4.045 | 3.436 | 3.165 | | 10 | gujarati | 3,789 | 7.4 | 4.079 | 3.489 | 3.226 | | 11 | argentinian_es | 4,362 | 6.8 | 4.077 | 3.520 | 3.246 | | 12 | colombian_es | 4,105 | 6.8 | 4.052 | 3.404 | 3.144 | | 13 | chilean_es | 3,755 | 6.6 | 4.056 | 3.424 | 3.160 | | 14 | tamil | 3,633 | 6.4 | 4.027 | 3.300 | 3.035 | | 15 | nigerian_en | 2,687 | 5.1 | 4.102 | 3.548 | 3.299 | | 16 | javanese | 2,906 | 4.2 | 4.004 | 3.467 | 3.163 | | 17 | malayalam | 2,470 | 4.1 | 4.034 | 3.367 | 3.095 | | 18 | aishell3 | 2,814 | 4.1 | 4.047 | 3.513 | 3.218 | | 19 | telugu | 2,517 | 4.0 | 4.053 | 3.358 | 3.098 | | 20 | venezuelan_es | 2,466 | 4.0 | 4.057 | 3.473 | 3.208 | | 21 | burmese | 2,071 | 3.6 | 4.020 | 3.495 | 3.203 | | 22 | sundanese | 2,116 | 3.5 | 3.999 | 3.443 | 3.140 | | 23 | khmer | 2,098 | 3.2 | 4.092 | 3.448 | 3.202 | | 24 | marathi | 1,400 | 2.8 | 4.088 | 3.433 | 3.186 | | 25 | yoruba | 1,464 | 2.2 | 4.035 | 3.476 | 3.178 | | 26 | nepali | 1,157 | 2.0 | 4.000 | 3.496 | 3.183 | | 27 | puertorico_es | 516 | 0.9 | 4.050 | 3.443 | 3.169 | | 28 | cv_ta | 263 | 0.4 | 3.855 | 3.116 | 2.780 | ## Schema Each config is a single `train` split with columns: | column | type | description | |---|---|---| | `audio` | `Audio(sampling_rate=48000)` | mono waveform, decoded to 48 kHz | | `source` | `string` | source/config name | | `bak` | `float32` | DNSMOS P.835 background-noise MOS (≥ 3.644 for every row) | | `sig` | `float32` | DNSMOS P.835 signal MOS | | `ovrl` | `float32` | DNSMOS P.835 overall MOS | | `duration` | `float32` | clip length in seconds | Long recordings are chunked to ≤ 15 s; short utterances (≥ 4 s) are kept whole. ## Usage ```python from datasets import load_dataset # load one config (source) ds = load_dataset("Scicom-intl/TTS-Clean44k", "uk_ireland_en", split="train") print(ds[0]["audio"], ds[0]["bak"], ds[0]["duration"]) # or stream a large config ds = load_dataset("Scicom-intl/TTS-Clean44k", "hifitts", split="train", streaming=True) ``` ## Sources - **OpenSLR high-quality TTS** — Javanese (41), Sundanese (44), Tamil (65), Telugu (66), Malayalam (63), Marathi (64), Khmer (42), Nepali (43), Gujarati (78), Kannada (79), Burmese (80). - **OpenSLR crowdsourced** — Argentinian/Chilean/Colombian/Peruvian/Puerto-Rican/Venezuelan Spanish (61/71/72/73/74/75), Catalan (69), Basque (76), Galician (77), Yoruba (86), Nigerian English (70), South-African English (32), UK & Ireland English (83). - **AISHELL-3** (Mandarin), **Hi-Fi TTS** (English, capped at 30 h for balance), **Common Voice** (Mandarin `zh-CN`, Tamil `ta`). ## What the gates rejected The verification is strict on purpose — sources that only *claim* to be high-fidelity were dropped: - **Common Voice `id` (Indonesian)** and **`yue` (Cantonese)** — excluded entirely; every clip was below 44.1 kHz (Cantonese was uniformly 32 kHz). - **VCTK** — omitted here only because its available mirror was download-throttled, not for quality. Nothing below the sample-rate or DNSMOS bar is included. ## Licensing Audio is redistributed from the upstream corpora listed above; each retains its **original license** (OpenSLR corpora are variously CC BY / CC BY-SA / CC0, AISHELL-3 is research-use, Hi-Fi TTS is CC BY 4.0, Common Voice is CC0). Consult the corresponding source before commercial use. The DNSMOS scores and the 48 kHz re-encoding are provided as-is. ## `labeled44k` — 2.87M transcribed clips, ≥44.1 kHz (added 2026-08) The other 32 subsets are a DNSMOS-filtered clean-TTS pool and carry **no transcripts**; their `audio.path` is also empty, which makes a row impossible to trace back to its upstream metadata. `labeled44k` is a separate, much larger pool built to close both gaps. | | | |---|--:| | clips | **2,870,194** | | duration | **4,297 h** | | size | 150.3 GB across 361 parquet shards | | sample rate | 1,683,440 @ 48 kHz + 1,186,754 @ 44.1 kHz (**per-clip verified ≥ 44.1 kHz**) | | transcripts | **every clip** (`text`) | | speaker labels | **every clip** (`speaker`), 206,102 distinct | Provenance: assembled from [`malaysia-ai/Multilingual-TTS`](https://huggingface.co/datasets/malaysia-ai/Multilingual-TTS), joining each source's `/train-*.parquet` (`audio_filename` → `text`) against the matching `_audio.zip`, across 317 sources. Audio bytes are the **original upstream encoding**, not re-encoded. Columns: `audio{bytes,path}`, `source`, `duration`, `sr`, `lang`, `speaker`, `text`. **Speaker labels cover 100% of clips** (2,870,194/2,870,194; **206,102 distinct speakers**), recovered from the upstream `speaker` column. The join is exact rather than fuzzy: each file is named `{source}__{md5(upstream_filename)[:12]}`, so the md5 maps every clip back to its upstream row deterministically — a clip either matches or is reported, never guessed. Note speaker ids are namespaced per source (e.g. `hifi-tts_9017`), so they are unique across the corpus and safe to group on directly. Two caveats worth reading before you use it: - **`audio` declares no fixed `sampling_rate`**, unlike the other subsets, because this pool is genuinely mixed. Read the real rate from `sr`; if you cast the column to a fixed rate you will resample 41% of the corpus. - **`lang` is unreliable: 1,795,530 clips (62.6%) are `unk`.** The tag comes from keyword-matching the source name, and most of the 317 sources match no keyword; only 23 distinct tags exist. The `text` itself is trustworthy and its script identifies the language — prefer script detection over this column. No DNSMOS (`bak`/`sig`/`ovrl`) is provided for this subset: it was not measured for this pool, and the fields are left absent rather than filled with placeholder values. Unlike the curated subsets, `labeled44k` is **not** DNSMOS-filtered — it is filtered on sample rate, decodability, duration and transcript presence only.