FlexiSLM/FlexiSLM-Data-4M-s2s

FlexiSLM-Data — Speech-to-Speech Part (4M)

• Paper: https://arxiv.org/abs/2606.31247
• Demo page: https://flexislm.github.io/
• Code: https://github.com/AmphionTeam/FlexiSLM
• Text (t2t) release: https://huggingface.co/datasets/FlexiSLM/FlexiSLM-Data-5M-t2t

FlexiSLM-Data is a large-scale, single-turn English speech-to-speech (s2s) dialogue dataset built to train FlexiSLM, a spoken language model. It is used in Stages 2 and 3 of FlexiSLM training, where s2s dialogue is the dominant task, alongside TTS (Emilia, MLS), ASR (MLS, LibriSpeech), audio understanding (LLaSO-Instruct), and text-to-text dialogue (TriviaQA, WebQuestions).

This repository contains the speech (s2s) part of FlexiSLM-Data: paired user-prompt and assistant-response audio in WebDataset format, before quality filtering. The underlying text pairs are released separately in FlexiSLM/FlexiSLM-Data-5M-t2t; samples share the same uuid, so the two releases can be joined directly.

Note: this release is a reproduction of the FlexiSLM-Data construction pipeline described in the paper. Dataset statistics are computed from the released files and may therefore differ from the numbers reported in the paper.

At a glance: the prompts are aggregated from 10 public QA, instruction-following, and dialogue datasets (TriviaQA, WebQuestions, TyDiQA, Alpaca, SmolTalk2, SODA, Magpie-Pro, UltraChat, HH-RLHF, WildChat) and voiced with Fish-Audio TTS using speakers sampled from English Emilia; every response is generated by Qwen3-Omni-30B-A3B and voiced with Qwen3-TTS using the fixed speaker "Ryan".

Construction pipeline

1. Prompt collection. Text prompts are collected from public QA, instruction-following, and dialogue datasets (see the table below). For multi-turn datasets, the user's first-turn utterance is used as the prompt; samples whose first turn is a generic greeting (e.g. "Hello") are skipped.
2. Response generation. All text responses are generated with Qwen3-Omni-30B-A3B; the responses shipped with the source datasets are discarded. We use a spoken language model rather than a text-only LLM because Qwen3-Omni produces speech-friendly responses: short, conversational, and free of formatting that does not transfer to speech (bullet points, code blocks, long enumerations). The text version of the prompt is fed to the model rather than its synthesized speech, since text-input responses are typically more accurate.
3. Speech synthesis. Responses are synthesized with Qwen3-TTS using the fixed speaker "Ryan". Prompts are synthesized with Fish-Audio TTS, with speaker prompts randomly sampled from English Emilia utterances longer than 5 seconds. This yields the 4,222,459 samples and 26,735 hours of audio shipped here.
4. Quality filtering (applied downstream of this release). Format-based filtering removes samples containing code, formulas, excessive punctuation, or non-target languages; correctness filtering uses the DeepSeek-V4-Flash API; ASR-based filtering with Whisper-medium discards cases with WER > 20%. The filtered training subset used for FlexiSLM totals 9.9K hours of speech (2.7K hours of user prompts and ~7.2K hours of assistant responses).

Prompt sources

Only the user prompts come from these datasets; their original answers are not used.

Dataset	Type	# Prompts
TriviaQA	QA	138K
WebQuestions	QA	3.8K
TyDiQA	Multilingual QA	167K
Alpaca	Instruction	52K
SmolTalk2	Instruction / Dialogue	385K
SODA	Dialogue	1.48M
Magpie-Pro	Instruction / Dialogue	1M
UltraChat	Multi-turn Dialogue	949K
HH-RLHF	Dialogue / Preference	167K
WildChat	Real-user Dialogue	159K

Statistics

	value
Samples (train)	4,216,459 across 1,406 shards
Samples (validation)	6,000 across 2 shards
Samples (total)	4,222,459
Prompt audio	7,317.3 hours (26,342,351 s)
Response audio	19,418.2 hours (69,905,434 s)
Total audio	26,735.5 hours (96,247,785 s)
Mean prompt duration	6.24 s
Mean response duration	16.56 s

Layout

Each sample is three tar members sharing an eight-digit key, unique across every shard and both splits:

00000001.question.wav   # user prompt, 44.1 kHz 16-bit mono, Fish-Audio TTS
00000001.response.mp3   # assistant reply, Qwen3-TTS speaker "Ryan"
00000001.json           # transcripts and per-sample metadata

• train: 4216459 samples across 1406 shards (shards/train-{00000..01405}-of-01406.tar)
• validation: 6000 samples across 2 shards (shards/validation-{00000..00001}-of-00002.tar)

Exact per-shard sample counts and byte offsets live in shards.json; per-sample records live in manifest.jsonl.

Each metadata record carries:

field	description
key	Eight-digit WebDataset key
uuid	Stable identifier, shared with the t2t release
split	train or validation
shard	Shard the sample lives in
question_text / response_text	Transcripts of the two audio members
question_duration / response_duration	Seconds
total_audio_duration	Seconds
num_tokens_est	Estimated audio tokens at 12 tokens/s

Loading

from datasets import load_dataset

dataset = load_dataset("FlexiSLM/FlexiSLM-Data-4M-s2s", split="train", streaming=True)

Or straight from WebDataset, using the brace pattern recorded in shards.json:

import webdataset as wds

url = "https://huggingface.co/datasets/FlexiSLM/FlexiSLM-Data-4M-s2s/resolve/main/shards/train-{00000..00099}-of-01406.tar"
dataset = wds.WebDataset(f"pipe:curl -sL {url}")

Limitations

• Single-turn only. Multi-turn sources are truncated to their first user turn.
• Unfiltered. These are the raw synthesized pairs before the format, correctness, and ASR-based filtering described above; some samples contain artifacts, mismatched audio, or non-English text.
• Synthetic speech. All audio is TTS-generated, with a single fixed response voice; it does not reflect real conversational acoustics, noise, or speaker diversity on the response side.
• Model-generated text. Every response comes from Qwen3-Omni-30B-A3B and inherits its biases and factual errors; none are human-verified.
• Derived data. Prompts inherit the licenses and terms of their source datasets; please check each source before redistribution.

References

• Qwen3-Omni — Xu et al., 2025
• Qwen3-TTS — Hu et al., 2026
• Fish-Audio TTS — Liao et al., 2024
• Emilia — He et al., 2024
• MLS — Pratap et al., 2020
• LibriSpeech — Panayotov et al., 2015
• LLaSO-Instruct — Sun et al., 2025
• TriviaQA — Joshi et al., 2017
• WebQuestions — Berant et al., 2013
• TyDiQA — Clark et al., 2020
• Alpaca — Taori et al., 2023
• SODA — Kim et al., 2023
• Magpie — Xu et al., 2025
• UltraChat — Ding et al., 2023
• HH-RLHF — Bai et al., 2022
• WildChat — Zhao et al., 2024