What is speech-expert?

A guide to speech-mcp, an MCP server and web app for turning text into speech, speech into text, and handling live voice interactions. It covers multiple speech providers, including offline and cloud options.

Use it for audio-file transcription, streaming speech recognition, text-to-speech, wake words, barge-in handling, realtime voice sessions, and voice command routing.

It helps choose a speech method that fits the job, such as file transcription, live partial captions, realtime conversation, or user interruptions.

speech-expert — Skill for Claude Code by sandraschi/speech-mcp

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent

npx skills add sandraschi/speech-mcp --skill speech-expert

Clone the repo

git clone --depth 1 https://github.com/sandraschi/speech-mcp

Made for: Claude Code.

Or install speech-mcp, the plugin that ships this one along with the rest of its 2 skills.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

README.md

[![agentmods](https://agentmods.dev/badge/skills/sandraschi/speech-mcp/speech-expert/github.svg)](https://agentmods.dev/skills/sandraschi/speech-mcp/speech-expert)

Your own site

<a href="https://agentmods.dev/skills/sandraschi/speech-mcp/speech-expert"><img src="https://agentmods.dev/badge/skills/sandraschi/speech-mcp/speech-expert/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

Your own site · 80×15

<a href="https://agentmods.dev/skills/sandraschi/speech-mcp/speech-expert"><img src="https://agentmods.dev/badge/skills/sandraschi/speech-mcp/speech-expert.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 36 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.

When invoked 919 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.

Origin original No closer match found in the catalogue.

Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

Model	Per session	Once invoked
Fable 5.1	$0.00036	$0.00919
Opus 5.5	$0.00014	$0.00368
Sonnet 5.5	$0.00007	$0.00184
Haiku 4.5	$0.00004	$0.00092

Measured 1mo ago against content hash 33e4d0c5f47a, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

speech-expert scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

src/speech_mcp/skills/speech-expert/SKILL.md · 63 lines

How it starts

The opening of the file, as written. The whole thing — 63 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Speech-MCP Expert

speech-mcp is a multi-provider speech gateway: text-to-speech (TTS), speech-to-text (STT), realtime voice, wake word, and a fleet voice command bus. It runs as an MCP server (stdio or HTTP) with a React webapp.

When to use which STT

Need	Use	Why
Batch/file transcription	`transcribe_audio_file` (provider `funasr`)	FunASR MLT-Nano: offline, multilingual, VAD+punctuation+diarization
Live streaming with partials	`streaming_stt` (sherpa-onnx)	ja/en/de, CPU, emits partials + endpoint
Barge-in / interruption	`barge_in_feed`	Completed utterance = interrupt signal
Cloud quality, realtime conversation	`start_evi_session` / Gemini Live	Hume EVI or `gemini_live`
Tiny chunks / stateless	`transcribe_stream_chunk`	base64 PCM per fragment

Guidance: default to funasr for accuracy + offline; use streaming_stt when you need text as it is spoken (voice agents, live captions); use barge_in_feed to let the user interrupt the assistant (no echo cancellation - gate mic feeding while it speaks).

TTS providers

text_to_speech(provider=...) - windows (SAPI5, always works), gemini (highest cloud quality, voice=Kore etc.), hume (emotional/expressive), elevenlabs (needs voice_id from manage_voice_clones), gemma (local, SAPI fallback). text_to_dialogue = multi-voice ElevenLabs conversation.

Wake word + voice command bus

configure_local_wake_word(action="start|stop|status") - openWakeWord, offline.
With FLEET_VOICE_DELEGATE=1 the listener routes wake -> STT -> POST fleet-agent /api/voice/intent. SHERPA_ASR_ENABLED=1 switches capture to streaming (no fixed window) with a follow-up turn.

REST surface

/api/v1/health (provider booleans), /api/v1/diagnostics, /api/tools, /api/skills, /api/v1/tts, /api/v1/voices, /api/v1/transcribe, /api/v1/wake_word, /api/v1/utility, /api/v1/stop, /api/v1/shutdown.

Config essentials

FunASR: FUNASR_ENABLED=true, FUNASR_MODEL=FunAudioLLM/Fun-ASR-MLT-Nano-2512 (31 langs incl. en/ja/de), FUNASR_DEVICE=cuda:0|cpu. Requires uv sync --extra funasr.
Streaming: SHERPA_ASR_ENABLED=true, SHERPA_ASR_LANG=en|ja|de, SHERPA_BARGE_IN=true. Requires uv sync --extra sherpa + just sherpa-download.
Cloud keys: GOOGLE_API_KEY, HUME_API_KEY, ELEVENLABS_API_KEY.

Read the full file on GitHub · 63 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

1mo ago First seen · 63 lines · 36 tokens per session scan A 33e4d0c5f47a

Subscribe to this mod's changes

speech-expert is a skill published in the GitHub repository sandraschi/speech-mcp (2 stars, last pushed yesterday), licensed MIT. It adds 36 tokens to every session and 919 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Creative and media elevenlabs fastmcp funasr mcp speech-to-text text-to-speech

ai-media

Use when a creative goal must become a finished media file: pick and order generative-media models per modality — AI voiceover, image-to-video clips, score — then glue them with ffmpeg (mux, duck, loudnorm, concat). NOT still-image generation/editing (that is replicate-images); NOT code-rendered React compositing…

ericrisco/rsc-harness · 82 tokens

speak

Provides concise spoken alerts and digests through the mcp-tts MCP server. It should be selected proactively for significant AI and DevOps transitions: a substantial plan is ready, a long build/test/deploy/release/monitoring phase changes state, a major task completes, a terminal blocker or exhausted retry budget…

blacktop/mcp-tts · 99 tokens

start-talking

Start and run an explicit, interactive Cadence Code conversation with Codex or Antigravity using fully local speech input and output. Use only when the user explicitly invokes $start-talking, /start-talking, or asks to start talking with Cadence Code.

michael-L-i/cadence-code · 57 tokens

speech-expert