Speech AI (Pronunciation + TTS + STT)

What is Speech AI (Pronunciation + TTS + STT)?

/mcp/pronunciation/mcp is an MCP server that exposes pronunciation assessment, text-to-speech, and speech-to-text through a single Streamable HTTP endpoint. It uses compact ONNX models for low-latency inference (<500 ms for pronunciation, <1 s for TTS) and supports English for STT and pronunciation, with American and British voices for TTS. Suitable for AI agents needing real‑time speech capabilities.

How to use Speech AI (Pronunciation + TTS + STT)?

An API key is required; obtain it from the Brainiall portal (GitHub sign‑in, purchase credits, create key). Add the key as Ocp-Apim-Subscription-Key (or Authorization/api-key) in the header of every request. Configure the MCP server URL (https://apim-ai-apis.azure-api.net/mcp/pronunciation/mcp) with that header in Claude Desktop, Cursor, or Cline (example JSON config is provided). Use the server’s 10 tools, 8 resources, and 3 prompts to perform speech tasks.

Key features of Speech AI (Pronunciation + TTS + STT)

• Pronunciation assessment with per‑phoneme scoring (39 ARPAbet)
• Text‑to‑speech with 12 American/British voices, 24 kHz WAV
• Speech‑to‑text with compact 17 MB model, word timestamps
• Pricing from $0.01 per STT request to $0.03 per 1K TTS chars
• Streamable HTTP transport for AI agents (no WebSocket)
• Available on Smithery (score 95/100), MCPize, Apify, MCP Registry

Use cases of Speech AI (Pronunciation + TTS + STT)

• Interactive pronunciation tutors with real‑time feedback
• Voice‑enabled AI assistants that transcribe and speak responses
• Automated language learning tools scoring spoken phrases
• Real‑time captioning or transcription from audio streams

FAQ from Speech AI (Pronunciation + TTS + STT)

What speech tasks does the server support?

It offers pronunciation assessment (score 0–100 with phoneme‑level details), text‑to‑speech synthesis (12 voices), and speech‑to‑text transcription (English, with word timestamps). All three are exposed as MCP tools.

How do I authenticate with the server?

Include one of these headers in every request: Ocp-Apim-Subscription-Key, Authorization: Bearer YOUR_KEY, or api-key. The API key is generated from the Brainiall portal after signing in with GitHub and purchasing credits.

Where are the API keys managed and is my data processed?

API keys are managed in the Brainiall portal (credit packages $5–$100). The README does not specify data residency; requests are sent to the Azure‑hosted endpoint https://apim-ai-apis.azure-api.net. No on‑premises deployment is mentioned.

What are the runtime dependencies?

No local model download is needed – all inference runs on Brainiall’s cloud servers. The server only requires HTTPS connectivity and an API key. Client‑side examples use Python (requests, base64) and Node.js (openai SDK for LLM, but speech calls use raw HTTP).

Are there any known limitations?

The README does not list explicit limits beyond latency and pricing. English is supported for STT and pronunciation; TTS offers American/British voices only. The compact models (<17 MB) may have accuracy trade‑offs compared to larger models, but no benchmark data is provided.