VoxMel
最近关注到一个挺有意思的开源 TTS 项目: VoxMel 。 GitHub: https://github.com/VoxDance-Lab/VoxMel Twitter: https://x.com/_akhaliq/status/2120749548851368358 现在不少自回归语音生成模型会依赖学习式声学 Tokenizer,先把音频离散化再进行建模。 VoxMel 的思路有点不一样,它尝试直接在连续 Mel 频谱上做自回归生成,不依赖学习式声学 Tokenizer,同时通过小波分解来组织不同尺度的声学信息。 感觉这个方向还挺值得探索的,毕竟如何在不引入离散声学 Token 的情况下实现高质量语音生成,一直是个挺有意思的问题。 项目已经开源了代码、模型和技术报告,也有音频 Demo 可以试听。 分享
This explanation quotes or summarizes the cited source record. Saving creates a bookmark on this device, with no automatic account sync.
- First observed in our collection
- Nov 16, 2026, 11:33 PM UTC
- Earliest cited publication
- Nov 15, 2026, 12:00 AM UTC
- Last observed
- Nov 17, 2026, 12:02 PM UTC
Evidence in the last complete seven days
- Source mentions
- 2 Previous seven days 0
- Distinct sources
- 2 Previous 0
- Comparable growth
- Not comparable
- Ranking appearances
- 0
Collection observations and undated records are separate from dated publication counts. Observed in 7d:1 · Undated records:0
Collection coverage · Comparable windowsCurrent / previous observed days 7 / 7
Counts use complete UTC days. First observed means first seen by this collection, not when a term was coined.
Current window:2026-11-10 → 2026-11-17 (end excluded) · History 29 days
| Source | Current days | Previous days | Last success |
|---|---|---|---|
| arXiv | 6 | 7 | Nov 16, 2026, 1:19 PM UTC |
| arXiv | 7 | 7 | Nov 16, 2026, 1:19 PM UTC |
| Ask HN | 7 | 7 | Nov 17, 2026, 12:01 PM UTC |
| Ask HN | 7 | 7 | Nov 16, 2026, 11:33 PM UTC |
| Ask HN | 7 | 7 | Nov 16, 2026, 11:33 PM UTC |
| Ask HN | 7 | 7 | Nov 16, 2026, 11:33 PM UTC |
| Digg | 1 | 0 | Nov 17, 2026, 12:01 PM UTC |
| GitHub | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| GitHub Issues | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| GitHub Releases | 7 | 7 | Nov 17, 2026, 5:46 AM UTC |
| GitHub | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| GitHub Issues | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| GitHub Releases | 7 | 7 | Nov 17, 2026, 5:46 AM UTC |
| GitHub | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| GitHub Issues | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| GitHub Releases | 7 | 7 | Nov 17, 2026, 5:46 AM UTC |
| GitHub Discovery | 7 | 5 | Nov 17, 2026, 12:02 PM UTC |
| GitHub Trending | 7 | 7 | Nov 17, 2026, 12:01 PM UTC |
| 7 | 7 | Nov 16, 2026, 1:19 PM UTC | |
| Hackobar | 1 | 0 | Nov 17, 2026, 12:02 PM UTC |
| Hugging Face | 4 | 7 | Nov 16, 2026, 1:19 PM UTC |
| Hugging Face Spaces | 5 | 5 | Nov 17, 2026, 12:01 PM UTC |
| Hacker News | 7 | 7 | Nov 17, 2026, 12:01 PM UTC |
| Hacker News | 7 | 7 | Nov 16, 2026, 11:33 PM UTC |
| Hacker News | 7 | 7 | Nov 16, 2026, 11:34 PM UTC |
| Hacker News | 7 | 5 | Nov 17, 2026, 12:02 PM UTC |
| Show HN | 7 | 5 | Nov 17, 2026, 5:46 AM UTC |
| Microlaunch | 1 | 0 | Nov 16, 2026, 11:34 PM UTC |
| MIT News | 7 | 6 | Nov 16, 2026, 1:19 PM UTC |
| OpenAI | 7 | 7 | Nov 16, 2026, 1:19 PM UTC |
| Product Hunt | 7 | 7 | Nov 17, 2026, 12:02 PM UTC |
| 量子位 | 7 | 5 | Nov 16, 2026, 1:19 PM UTC |
| OpenRouter | 7 | 5 | Nov 17, 2026, 5:46 AM UTC |
| GitHub Trending | 3 | 7 | Nov 12, 2026, 4:00 AM UTC |
| Hugging Face Models | 5 | 5 | Nov 17, 2026, 12:02 PM UTC |
| HF Daily Papers | 5 | 4 | Nov 17, 2026, 12:02 PM UTC |
| OpenRouter | 7 | 5 | Nov 17, 2026, 5:46 AM UTC |
| Product Hunt | 3 | 7 | Nov 12, 2026, 4:00 AM UTC |
| SEC EDGAR | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| Sequoia Capital | 7 | 7 | Nov 16, 2026, 1:19 PM UTC |
| Simon Willison | 7 | 5 | Nov 16, 2026, 1:19 PM UTC |
| TechCrunch | 7 | 7 | Nov 17, 2026, 12:01 PM UTC |
| Techmeme | 1 | 0 | Nov 17, 2026, 12:02 PM UTC |
| Trendshift | 1 | 0 | Nov 17, 2026, 12:02 PM UTC |
| Contracts Finder | 7 | 7 | Nov 17, 2026, 5:45 AM UTC |
| Uneed | 1 | 0 | Nov 17, 2026, 12:02 PM UTC |
| V2EX | 5 | 7 | Nov 17, 2026, 12:01 PM UTC |
- Counts describe monitored public sources, not the whole web or search demand.
Daily history · up to 90 days
Bars show dated source mentions. The latest UTC day may be incomplete. Zero mentions and missing collection coverage are different conditions.
View accessible daily data
| UTC date | Mentions | Sources | Observations | Ranking appearances |
|---|---|---|---|---|
| 2026-11-03 | 0 | 0 | 0 | 0 |
| 2026-11-04 | 0 | 0 | 0 | 0 |
| 2026-11-05 | 0 | 0 | 0 | 0 |
| 2026-11-06 | 0 | 0 | 0 | 0 |
| 2026-11-07 | 0 | 0 | 0 | 0 |
| 2026-11-08 | 0 | 0 | 0 | 0 |
| 2026-11-09 | 0 | 0 | 0 | 0 |
| 2026-11-10 | 0 | 0 | 0 | 0 |
| 2026-11-11 | 0 | 0 | 0 | 0 |
| 2026-11-12 | 0 | 0 | 0 | 0 |
| 2026-11-13 | 0 | 0 | 0 | 0 |
| 2026-11-14 | 0 | 0 | 0 | 0 |
| 2026-11-15 | 1 | 1 | 0 | 0 |
| 2026-11-16 | 1 | 1 | 1 | 0 |
| 2026-11-17 | 0 | 0 | 1 | 1 |
Source evidence
Sources in the evidence shown:V2EX 1 · HF Daily Papers 1
V2EX · Published Nov 16, 2026
[程序员] [开源分享] 发现一个挺有意思的 TTS 项目 VoxMel,直接在 Mel 频谱上做自回归生成
最近关注到一个挺有意思的开源 TTS 项目: VoxMel 。 GitHub: https://github.com/VoxDance-Lab/VoxMel Twitter: https://x.com/_akhaliq/status/2120749548851368358 现在不少自回归语音生成模型会依赖学习式声学 Tokenizer,先把音频离散化再进行建模。 VoxMel 的思路有点不一样,它尝试直接在连续 Mel 频谱上做自回归生成,不依赖学习式声学 Tokenizer,同时通过小波分解来组织不同尺度的声学信息。 感觉这个方向还挺值得探索的,毕竟如何在不引入离散声学 Token 的情况下实现高质量语音生成,一直是个挺有意思的问题。 项目已经开源了代码、模型和技术报告,也有音频 Demo 可以试听。 分享
View source record →HF Daily Papers · Published Nov 15, 2026
VoxMel TTS Technical Report
Continuous autoregressive TTS relies on an acoustic target that enables stable next-latent modeling while preserving fine-grained acoustic details. While existing approaches employ learned compressed representations (e.g., continuous VAE latents) to render prediction tractable, they rely on complex neural encoders that often cap generation fidelity. Converse
Review priority 52 / 100
This score orders research candidates. It is not search volume, a forecast or verified commercial demand.
Evidence
12.7 / 30Distinct dated articles or first observations; repeat crawling adds nothing.
Publisher diversity
16 / 25Conservative owner grouping; syndicated copies count once.
Comparable change
0 / 20Comparable complete weeks only; no percentage from a zero baseline.
Recent evidence
15 / 15Latest publication or first observation, not repeated fetch time.
Name specificity
7.8 / 10Explicit entities and technical concepts outrank ambiguous acronyms.
Limitations and interpretation
- First observation is a collection milestone, not proof that a word or product is new to the world.
- Counts describe monitored public sources, not the whole web or search demand.
- First observed here is not the invention date of a name or idea.
- Candidates come from public titles, summaries and entity URLs, not verified full-text analysis.
- Owner grouping is conservative; undisclosed syndication may remain.
- Repeat ranking appearances are observations, not new articles or independent endorsements.
- The previous week is too small for a reliable percentage change.