Early-Signal Detector

task · first seen 4 Oct 2026

text-to-speech

Also: TTS, speech synthesis, expressive speech synthesis

Generating spoken audio from written text using neural models.

Text-to-speech is the long-standing task of converting written text into spoken audio, now dominated by neural approaches. The sources show ordinary activity across the field, with no single new idea driving it: an interpretability paper tracing emotion-control circuits in LLM-based TTS, a Yorùbá text corpus release, Turkish models distilled from VoxCPM teachers, a study of classifier-free guidance hurting speaker similarity in few-step flow-matching decoders, and Lombard-effect adaptation. On the code side there are on-device Core ML runtimes, tiny ONNX models, inclusionAI's Ming-3.0-tts MoE family, and Claude Code skills that use TTS or voice cloning as a component.

The field is mature, so the mention spike reflects broad background volume rather than a new technique. The narrower threads may be worth tracking separately: tiny models distilled entirely on synthetic speech, few-step flow-matching decoders, and LLM-based TTS with natural-language voice control.

Analyst note (established), 17 Nov 2026.

Mentions per week

01619 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 79 on 14 Nov 2026 · 48 mentions in 28 days · 45 independent groups · papers → code → discussion · first seen 6 weeks ago · acceleration -0.29

Earliest sources

Latest sources

Mentioned together with

voice cloningautomatic speech recognitionagent skillstext-to-video generationmodel context protocolWhisperzero-shot voice cloningself-hosted LLM deploymentmixture of expertsLombard effect