laion/tts-scaling-ladder-de-en

TTS Scaling Ladder DE/EN — eight nested, balanced, openly licensed speech tiers

A nested ladder of eight speech datasets for scaling-law experiments on TTS, voice-conversion and audio foundation models. Every tier is a strict subset of the next; every tier is 50/50 German/English, within each language 50/50 real/synthetic, and inside each of those four cells stratified to be uniform over 40 EmoNet emotion categories (pipeline A) and uniform over 57 VoiceNet voice dimensions × 10 deciles (pipeline B). The smallest tier is the most selective: it holds the most extreme emotions and the highest quality relative to the rest.

Every element ships audio (MP3, byte-identical to its source repository), MOSS-Audio-Tokenizer-v2 codes (uint16 [T, 12], 12.5 fps) and a JSON record with the procedural voice-acting prompt in the LAION-AI/procedural-voice-captions format (GENERAL line + SCRIPT with per-sentence delivery cues, [pause X.Xs] markers and inline (Vocal Burst) labels), plus the raw scores it was built from, the language, the origin, the source, the source uid, the source licence and the duration.

tier	hours per language (DE / EN)	total	purpose	elements in this release	hours in this release

| 0 | 50 / 50 | 100 | micro-core, baseline, hyper-parameter search | 41,771 | 100.0 | | 1 | 125 / 125 | 250 | small convergence test | 100,858 | 249.9 | | 2 | 250 / 250 | 500 | saturation of acoustic features | 197,662 | 500.1 | | 3 | 500 / 500 | 1,000 | compute-optimal anchor (1×) | 382,768 | 1,000.0 | | 4 | 1,250 / 1,250 | 2,500 | 2.5× | 912,789 | 2,500.1 | | 5 | 2,500 / 2,500 | 5,000 | 5× | 1,764,990 | 5,000.4 | | 6 | 5,000 / 5,000 | 10,000 | 10× | 3,425,099 | 10,000.7 | | 7 | 10,000 / 10,000 | 20,000 | full-corpus benchmark, 20× | 6,666,998 | 20,001.5 | | held-out | 40 / 40 | 80 | evaluation split, disjoint from every tier, same stratification | 25,639 | 80.1 |

Held-out split. 20 h per cell (10 h per pipeline) drawn with exactly the same rules as the tiers — the next elements behind the tier-7 boundary of every bucket/category in master-rank order, same normalisation, same quotas — and disjoint from all tiers (tier = 8, verified by uid: zero overlap). It lives under data/*/*/heldout/ and index/*/*/heldout/, loader config heldout. Use it for the scaling-law loss; never train on it.

Hours are summed over dur_s. Tier N ⊂ Tier N+1 strictly — by construction, not by check: the whole release is stored once (the tier-7 stock plus held-out, 6,692,637 elements, 20,081.6 h, 2,694 shards, 1.04 TB of tars) with a column tier = the smallest tier index that contains the element, and the tierN loader configs simply filter tier <= N.

Two assumptions made by the executor (the specification left them open)

1. 50/50 split between the two pipelines. Inside every cell (language × origin) half of the volume is filled by pipeline A (uniform over 40 EmoNet categories) and half by pipeline B (uniform over 57 VoiceNet dimensions × 10 deciles). The specification fixed the strata but not the share; halves were the coordinator's assumption.
2. One stock, one tier column. The release is stored exactly once (the tier-7 stock plus the held-out split) and every element carries tier = the smallest tier that contains it. Tier K is therefore defined as tier <= K (and, on disk, as the tar directories tier0 … tierK); nesting is guaranteed by construction rather than by shipping eight overlapping copies.

The remaining executor's choices (pool definition, decile edges, deduplication order, bucket shortfall handling, language verdicts) are marked [A] in the rules below.

Loading — one example per tier

Every tierN config is the union of the tar/parquet directories tier0 … tierN; heldout is the evaluation split.

from datasets import load_dataset
R = "laion/tts-scaling-ladder-de-en"
t0 = load_dataset(R, "tier0")      #    100 h  (50 DE + 50 EN; 25 h real + 25 h synth per language)
t1 = load_dataset(R, "tier1")      #    250 h
t2 = load_dataset(R, "tier2")      #    500 h
t3 = load_dataset(R, "tier3")      #  1,000 h  compute-optimal anchor
t4 = load_dataset(R, "tier4")      #  2,500 h
t5 = load_dataset(R, "tier5")      #  5,000 h
t6 = load_dataset(R, "tier6")      # 10,000 h
t7 = load_dataset(R, "tier7")      # 20,000 h  everything
ev = load_dataset(R, "heldout")    #     80 h  evaluation, disjoint from every tier
de = load_dataset(R, "de")         # one language, all tiers: filter de["train"]["tier"] <= N yourself

Each row is one element (metadata, scores, prompt); audio and MOSS codes are in the tar with the same basename:

Audio and MOSS codes live in WebDataset tars next to the index, one tar per index parquet, same basename:

import io, json, tarfile, numpy as np
from huggingface_hub import hf_hub_download
row = idx["train"][0]                                       # any index row
p = hf_hub_download("laion/tts-scaling-ladder-de-en", f"data/{row['lang']}/{row['origin']}/tier{row['tier']}/{row['shard']}.tar", repo_type="dataset")
with tarfile.open(p) as tf:
    mp3  = tf.extractfile(f"{row['uid']}.mp3").read()               # bytes, decode with soundfile/torchaudio
    moss = np.load(io.BytesIO(tf.extractfile(f"{row['uid']}.moss.npy").read()))   # uint16 [T, 12]
    rec  = json.load(tf.extractfile(f"{row['uid']}.json"))          # prompt, words, sentences, bursts, scores
print(rec["prompt"])

Tar members per element: <uid>.mp3, <uid>.moss.npy, <uid>.json. Every tar of tier block tierN holds only elements whose smallest tier is N, so tier K = all tars under tier0 … tierK. For streaming with webdataset:

import webdataset as wds
K = 3
urls = [f"https://huggingface.co/datasets/laion/tts-scaling-ladder-de-en/resolve/main/data/{lang}/{origin}/tier{n}/{shard}.tar"
        for (lang, origin, n, shard) in shards_of_tier(K)]      # from the index: columns lang, origin, tier, shard
ds = wds.WebDataset(urls).decode().to_tuple("mp3", "moss.npy", "json")

Decoding note. The voice-profile sources (vprof_original, vprof_repair) are 48 kHz MP3 with two identical channels; decode with soundfile/torchaudio and take mean(axis=channels) — never reshape(-1) (that yields a half-speed signal of twice the length). All other sources are mono. audio_sr and audio_channels are in the index.

Selection rules (verbatim from the specification, with the executor's assumptions marked [A])

Licence filter (before any scoring)

Only openly licensed sources enter the pool. Real: emolia (DE+EN), mls (DE; the corpus has no English MLS), eurospeech (DE+EN, explicitly released by the project lead), kartoffelphon (DE), commonvoice (DE and EN subsets only). Synthetic: laion-voice-profiles-annotated (origin=original, origin=repair, vc_sidon — 500 LAION voice profiles, DE/EN in halves), laion/voice-acting-edge-top3 (formerly laion/dramabox-edge-top3-voice), laion/voice-acting-reinterpretations-top3 (formerly laion/dramabox-reinterpretations-top3) (MOSS takes). Excluded: podcast, snippets, evasnippets (research repo, non-commercial), everything Mediathek, everything without a documented licence, laion/voice-acting-burst-audio (no score columns), laion-emotional-trajectory-t80 (chains concatenated from vc_sidon clips — duplicates — and 93 % language-mixed), laion/moss-voice-identity-repairs (identical material to origin=repair, which carries the MOSS codes and scores).

Scoring — all metrics min-max normalised per language × origin cell

S_gen = Norm(genuineness_0_6), S_vb = Norm(blend_0_10), S_emo,e = Norm(emo_e) for the 40 EmoNet dimensions, S_vn,d = Norm(vn_d_reg) for the 57 VoiceNet dimensions. R_quality = S_gen + S_vb; R_emotion(e) = 3·S_emo,e + R_quality. [A] min/max are exact over the whole cell pool (all sources of the cell together).

Stratification

Per cell (language × origin) the cell volume is split in halves between two pipelines ([A], coordinator's assumption).

• Pipeline A — EmoNet, 40 categories. For category e: candidates sorted by R_emotion(e); the intensity pool [A] is defined as nested bands: band k (k = 0…7) = the top prefix whose hours reach m · q_A(k) with m = 3 (the pool covers the quota threefold), q_A(k) = pipeline-A volume of tier k / 40. Master rank in A = (band ascending, then R_quality descending). Tier 0 is therefore the best-quality third of the most extreme band; every larger tier adds less extreme emotion and again ranks strictly by quality inside. Quota per category = 1/40 of the pipeline-A volume. If a category's pool is exhausted by deduplication, the ranking continues below band 7 in R_emotion(e) order (band = 8, counted in the report).
• Pipeline B — VoiceNet, 57 dimensions × 10 deciles = 570 buckets. For each dimension the value range [min, max] of the cell is cut into 10 equal-width intervals ([A]: equal width, not quantiles); inside each bucket the master rank is strictly R_quality. Quota per bucket = 1/570 of the pipeline-B volume, so extreme values (very low voice, very young age) are not disadvantaged. [A] A bucket with fewer available hours than its quota takes everything it has; the shortfall is redistributed proportionally to the other deciles of the same dimension (then, if needed, across all buckets of the pipeline).
• Deduplication. An element counts once. [A] Fixed order: pipeline A first, then B; inside A the 40 categories in alphabetical order (all pools are equal-sized); inside B the 570 buckets in ascending order of available candidate hours (scarcest first, which protects the extreme deciles), ties by (dimension, decile). An element already taken is skipped by later buckets. Elements that would have fallen into more than one window are counted (n_candidacies).
• Nesting. Per bucket/category the deduplicated list is walked in master-rank order, hours accumulated, and each element receives tier = min{N : cum_h ≤ q(N) + dur/2} (rounding to the nearest quota); elements beyond q(7) are dropped. [A] Bucket rounding noise is then balanced at the tier boundaries by moving boundary elements (last of tier n ↔ first of tier n+1) until every (cell, pipeline, tier) is within 0.25 % of its target — the prefix property is preserved. Stored once with tier, rank_in_bucket, bucket_id, pipeline, emotion or vn_dim/decile, plus band (A) and n_candidacies.
• [A] Language = the corpus lang verdict (upstream label + ASR LID, lang_src == agreed on 99.6 %); for the DramaBox sources fastText lid.176 over the ASR transcript. Duplicate uids inside a source (Emolia ships some clips in two upstream archives) are dropped before selection, first occurrence kept. No duration filter beyond dur_s > 0.

Licences per source

source (source)	HF repository	origin	licence of the audio	note
emolia	laion/laion-tts-annotated-v1 (emolia)	real	CC-BY-4.0 (laion/Emolia)
commonvoice	laion/laion-tts-annotated-v1 (commonvoice)	real	CC-0 (Mozilla Common Voice)	DE + EN subsets only
eurospeech	laion/laion-tts-annotated-v1 (eurospeech)	real	other — parliament terms vary; the upstream card makes the user responsible for compliance	included on the project lead's explicit release
kartoffelphon	laion/laion-tts-annotated-v1 (kartoffelphon)	real	CC-BY-4.0 (~51 % LibriVox public domain)
mls	laion/laion-tts-annotated-v1 (mls)	real	CC-BY-4.0 (facebook/multilingual_librispeech)	German only
vprof_original	laion/laion-voice-profiles-annotated (origin=original)	synth	CC-BY-4.0; generator laion/moss-tts-local-transformer-4.55b-voice-acting-v2 (Apache-2.0)	500 synthetic voice profiles; 124 mediathek_* voices are model-reinterpreted speakers, not broadcast audio
vprof_repair	laion/laion-voice-profiles-annotated (origin=repair)	synth	CC-BY-4.0	re-dos of takes that failed an identity check; both versions are distinct audio and both are in the pool
vc_sidon	laion/laion-voice-profiles-annotated (vc_sidon)	synth	CC-BY-4.0; Chatterbox VC (MIT) + SIDON	voice-conversion twin of the same takes as vprof_original (same script and performance, different timbre); one winning candidate per take
dramabox_edge	laion/voice-acting-edge-top3 (formerly laion/dramabox-edge-top3-voice)	synth	CC-BY-4.0 (synthetic material from LAION's own generation; annotations Apache-2.0)	266 hand-written edge cases × 2,000 prompts, rendered with DramaBox TTS (Resemble AI weights), top-3 of 20 seeds per group; language per clip from the ASR transcript
dramabox_reint	laion/voice-acting-reinterpretations-top3 (formerly laion/dramabox-reinterpretations-top3) (takes)	synth	CC-BY-4.0 (synthetic material from LAION's own generation)	MOSS-voice-acting-v2 reinterpretations (top-3 of 64 candidates) of DramaBox acting prompts; the unannotated DramaBox originals of that repo are not included

The license column carries this per element. If your use case cannot accept other, filter license to CC-BY-4.0 / CC-0 (this removes eurospeech; the tiers stay nested but no longer exactly balanced).

Balance (hours, from the shipped index)

tier	de_real A	de_real B	en_real A	en_real B	de_synth A	de_synth B	en_synth A	en_synth B	DE	EN	real	synth
0	12.5	12.5	12.5	12.5	12.5	12.5	12.5	12.5	50.0	50.0	50.0	50.0
1	31.2	31.2	31.3	31.2	31.3	31.3	31.2	31.3	125.0	125.0	124.9	125.0
2	62.5	62.5	62.5	62.5	62.5	62.5	62.5	62.5	250.0	250.0	250.0	250.0
3	125.0	125.0	125.0	125.0	125.0	125.0	125.0	125.0	500.0	500.0	500.0	500.0
4	312.5	312.5	312.5	312.5	312.5	312.5	312.5	312.6	1,250.0	1,250.1	1,250.0	1,250.1
5	625.0	625.0	625.0	625.1	625.0	625.0	625.0	625.2	2,500.0	2,500.3	2,500.1	2,500.2
6	1,250.0	1,250.1	1,250.0	1,250.2	1,250.0	1,250.1	1,250.0	1,250.3	5,000.3	5,000.5	5,000.3	5,000.4
7	2,500.0	2,500.3	2,500.0	2,500.3	2,500.0	2,500.3	2,500.0	2,500.6	10,000.6	10,000.9	10,000.6	10,000.9
held-out	10.0	10.0	10.0	10.0	10.0	10.0	10.0	10.0	40.1	40.1	40.0	40.1

Targets per (cell, pipeline), tiers 0–7: 12.5 / 31.25 / 62.5 / 125 / 312.5 / 625 / 1,250 / 2,500 h. Largest deviation over the 64 tier values: 0.204 %; over the eight held-out values: 0.238 %. All are within 0.25 %.

Hours per source (tier 7 plus held-out)

source	DE real	EN real	DE synth	EN synth
commonvoice	170.0 (131,515)	69.6 (57,578)	—	—
dramabox_edge	—	—	156.3 (19,747)	2,547.5 (310,985)
dramabox_reint	—	—	52.5 (12,805)	61.4 (15,617)
emolia	2,091.4 (669,330)	4,915.0 (1,505,156)	—	—
eurospeech	608.1 (140,828)	32.8 (7,916)	—	—
kartoffelphon	2,116.6 (611,467)	2.9 (923)	—	—
mls	34.1 (7,889)	—	—	—
vc_sidon	—	—	2,050.7 (709,047)	690.4 (394,290)
vprof_original	—	—	2,218.8 (747,642)	1,338.1 (651,627)
vprof_repair	—	—	542.0 (351,342)	383.3 (346,933)

The prompt (what rec["prompt"] / index prompt contains)

Produced offline from the stored annotations with the captioner of LAION-AI/procedural-voice-captions at main (PR #4: 17-class x2 head and energy-based pause correction), tags template, deterministic seed = stable_hash(uid):

GENERAL: very warm, very soft-onset, …, middle-aged, low-register, very slow, pining, helplessness, semi-genuine, with-bursts
SCRIPT:
(very flat, very smooth, no interest, very fatigue, genuine) Huh. (Chuckle)
(very flat, very warm, helplessness, concern, semi-genuine) [pause 0.4s] What have we here?

• GENERAL — top-5 VoiceNet dimensions by |z|·reliability, top-3 emotions, genuineness, burst blend, always Age/Gender/Register/Tempo, z-scored against the bundled in-domain baseline.
• SCRIPT — one line per sentence: (delivery cue) sentence text. Sentences are cut at sentence-final punctuation of the aligned words. [A] The cue of every sentence is derived from the clip's global scores (the corpora carry no per-sentence scores; the record stores the sentence timings so per-sentence re-scoring can be dropped in later). Wording rotates deterministically per sentence.
• [pause X.Xs] — a gap ≥ 0.30 s between consecutive words or sentences, measured from the stored word timestamps after the energy-based word-end correction (each word's end is trimmed back to where its speech actually stops, computed from the decoded audio; end_span keeps the raw value). Word timestamps are the corpus alignments (words in the source JSON — MMS-FA for the LAION corpora, Parakeet for the DramaBox sources); Parakeet was not re-run.
• (Vocal Burst) — the kept locator/classifier-v2 spans of the source annotation, written at the word gap where they happen (BURST_LABEL_SOURCE=v2). prompt_x2 is the same prompt with the 17-class x2 head naming the spans (BURST_LABEL_SOURCE=x2, recall-tiered wording: exact class / Class? / family word / Vocal Burst) — present where the vb2_* re-annotation exists (3,689,357 of 6,692,637 elements: the five real sources, vc_sidon shards with a re-annotation, dramabox_reint, and dramabox_edge where its re-annotation had landed), null elsewhere.
• Totals in this release: 8,508,596 pause markers, 3,928,534 kept bursts.
• rec["sentences"], rec["words"], rec["bursts"] and rec["scores"] are the record augment.py needs to regenerate varied phrasings of the same content at training time (augment_script(rec, seed)).

Columns

See COLUMNS.md. Every index row carries the raw scores (genuineness_0_6, blend_0_10, 40 emo_*, 57 vn_*_reg and, where the source has them, 57 vn_*_bucket), lang, origin, source, uid (= source uid), hf_source, license, dur_s, the selection fields and the prompt.

Provenance and verification

Built on JUPITER (JSC) from the local copies of the source repositories; nothing was re-encoded — every MP3 is the byte-identical member of its source tar, every MOSS code array is the source's (laion-tts-annotated-v1, laion-voice-profiles-annotated) or, for the two DramaBox sources, newly computed with OpenMOSS-Team/MOSS-Audio-Tokenizer-v2 at 48 kHz, n_vq = 12, with the frame-rate canary moss_frames / (decoded_duration × 12.5) ∈ [0.97, 1.03] checked on every element (the packer refuses otherwise). Upload with upload_large_folder; every file verified byte-exact against the local copy via repo_info(files_metadata=True). Selection code, statistics and the full report: ladder.json in this repo; the working report is scaling_ladder.md in the project.

Citation

If you use this ladder, credit LAION and the upstream sources listed above.