Skip to content

Speech AI - Pronunciation, STT & TTS

by fasuizu-br · remote · no sign-in to list tools · v2.3.0 · published by GitHub user fasuizu-br

Pronunciation scoring, speech-to-text, and text-to-speech for language learning

Changes · 8
CompareRegistry entry

10 tools: 8 change things and 2 only read.

Scanned Oct 21, 2026

1 check matched something in the tool definitions. No finding is not a safety guarantee. How the checks work.

claude mcp add --transport http speech-ai https://apim-ai-apis.azure-api.net/mcp/pronunciation/mcp

Findings

Matches of fixed checks, with the text they matched.

1 warn
  • Warnsynthesize_speech

    The server marks "synthesize_speech" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

    Matched text

    annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":true,"openWorldHint":true}; The description's first verb is "generate": a word for a change.

    What this check looks for

Tools

Open a tool for its reasons and full description. Descriptions are the publisher's own words, shown as they are.

10 tools

Changes things 8Claude Code: ask
  • assess_pronunciationAssess English pronunciation quality from audio.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Assess English pronunciation quality from audio. Scores pronunciation at four levels: overall, sentence, word, and phoneme. Each score is 0-100. Phonemes are returned in both IPA and ARPAbet notation. Sub-300ms inference latency. Args: audio_base64: Base64-encoded audio data. Supports WAV, MP3, OGG, and WebM formats. text: The reference English text that the speaker was expected to read aloud. audio_format: Audio format hint — one of 'wav', 'mp3', 'ogg', 'webm'. Defaults to 'wav'. Returns: dict with keys: - overallScore (int 0-100): Overall pronunciation quality - sentenceScore (int 0-100): Sentence-level fluency and accuracy - words (list): Per-word scores, each containing: - word (str): The word - score (int 0-100): Word pronunciation score - phonemes (list): Per-phoneme scores with IPA/ARPAbet notation - decodedTranscript (str): What the model heard (ASR transcript) - transcript (str): Reference text - confidence (float 0-1): Scoring confidence - warnings (list[str]): Quality warnings if any - audioQuality (dict): Audio metrics (SNR, peak/RMS dB, etc.)

    Annotations the server declares

    • title: Assess Pronunciation
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • check_pronunciation_serviceCheck if the pronunciation assessment service is healthy and ready.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Check if the pronunciation assessment service is healthy and ready. Returns: dict with keys: - status (str): 'healthy' or error state - modelLoaded (bool): Whether the scoring model is loaded - version (str): API version

    Annotations the server declares

    • title: Check Pronunciation Service
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • check_stt_serviceCheck if the speech-to-text service is healthy and ready.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Check if the speech-to-text service is healthy and ready. Returns: dict with keys: - status (str): 'healthy' or error state - modelLoaded (bool): Whether the STT model is loaded - version (str): API version

    Annotations the server declares

    • title: Check STT Service
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • check_tts_serviceCheck if the text-to-speech service is healthy and ready.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Check if the text-to-speech service is healthy and ready. Returns: dict with keys: - status (str): 'healthy' or error state - modelLoaded (bool): Whether the TTS model is loaded - version (str): API version

    Annotations the server declares

    • title: Check TTS Service
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • check_whisper_serviceCheck if the Whisper STT Pro service is healthy and ready.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Check if the Whisper STT Pro service is healthy and ready. Returns: dict with keys: - status (str): 'healthy' or error state - modelLoaded (bool): Whether the Whisper model is loaded - diarizeLoaded (bool): Whether the diarization pipeline is loaded - version (str): API version - modelName (str): Whisper model name (e.g. 'large-v3-turbo')

    Annotations the server declares

    • title: Check Whisper Service
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • synthesize_speechGenerate natural speech audio from English text.1 finding

    Why it is in this group

    • The description's first verb is "generate": a word for a change.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.

    Findings

    • Warn

      The server marks "synthesize_speech" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

      Matched text

      annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":true,"openWorldHint":true}; The description's first verb is "generate": a word for a change.

      What this check looks for
    Description, as the server wrote it
    Generate natural speech audio from English text. Produces high-quality speech with 12 English voices. Returns base64-encoded WAV audio (16-bit PCM, 24kHz mono) along with metadata. Available voices: - af_heart (default), af_bella, af_nicole, af_sarah, af_sky (American female) - am_adam, am_michael (American male) - bf_emma, bf_isabella (British female) - bm_george, bm_lewis, bm_daniel (British male) Args: text: English text to synthesize (1-5000 characters). voice: Voice ID. See list above. Defaults to 'af_heart'. speed: Speed multiplier from 0.5 to 2.0 (default: 1.0). Returns: dict with keys: - audio_base64 (str): Base64-encoded WAV audio (16-bit PCM, 24kHz) - duration_ms (str): Audio duration in milliseconds - voice (str): Voice ID used - text_length (str): Input text character count - processing_ms (str): Synthesis time in milliseconds

    Annotations the server declares

    • title: Synthesize Speech
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • transcribe_audioTranscribe audio to text with word-level timestamps.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Transcribe audio to text with word-level timestamps. Converts spoken English audio into text with optional word-level timestamps and per-word confidence scores. Args: audio_base64: Base64-encoded audio data (WAV, MP3, OGG, FLAC, WebM). audio_format: Audio format hint. Auto-detected from magic bytes if omitted. include_timestamps: Whether to include word-level timing (default: true). Returns: dict with keys: - text (str): Full decoded transcript - words (list): Per-word results with timestamps, each containing: - word (str): The transcribed word - start (float): Start time in seconds - end (float): End time in seconds - confidence (float 0-1): Word-level confidence - audioDurationMs (int): Audio duration in milliseconds - metadata (dict): Processing time, audio length, model version - audioQuality (dict): Audio metrics (SNR, peak/RMS dB, etc.)

    Annotations the server declares

    • title: Transcribe Audio
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • transcribe_audio_proTranscribe audio with Whisper Large V3 Turbo — multilingual STT.No findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Transcribe audio with Whisper Large V3 Turbo — multilingual STT. Supports 99 languages with automatic language detection, word-level timestamps, per-word confidence scores, and optional speaker diarization (identifies who spoke each word). Best-in-class WER (~2%). Args: audio_base64: Base64-encoded audio (WAV, MP3, OGG, FLAC, WebM). language: Language code. Auto-detected if omitted. Supports 99 languages. diarize: Enable speaker diarization (default: false). When true, each word includes a speaker label (e.g. SPEAKER_00, SPEAKER_01). Returns: dict with keys: - text (str): Full decoded transcript - words (list): Per-word results with timestamps, each containing: - word (str), start (float), end (float), confidence (float 0-1) - speaker (str|null): Speaker label when diarize=true - speakers (dict|null): Speaker info with count and labels - audioDurationMs (int): Audio duration in milliseconds - metadata (dict): Processing time, language, languageProbability - audioQuality (dict): Audio metrics (SNR, peak/RMS dB, etc.)

    Annotations the server declares

    • title: Transcribe Audio Pro (Whisper)
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
Only reads 2Claude Code: allow
  • get_phoneme_inventoryGet the full phoneme inventory supported by the pronunciation scorer.No findings

    Why it is in this group

    • The name "get_phoneme_inventory" says "get": a word for reading.
    Description, as the server wrote it
    Get the full phoneme inventory supported by the pronunciation scorer. Returns a list of all English phonemes the engine can assess, including ARPAbet symbol, IPA equivalent, example word, and phoneme category (vowel, consonant, diphthong). Returns: list of dicts, each with keys: - arpabet (str): ARPAbet symbol (e.g. 'AA', 'TH') - ipa (str): IPA notation - example (str): Example word containing the phoneme - category (str): vowel, consonant, or diphthong

    Annotations the server declares

    • title: Get Phoneme Inventory
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_tts_voicesList all available text-to-speech voices with metadata.No findings

    Why it is in this group

    • The name "list_tts_voices" says "list": a word for reading.
    Description, as the server wrote it
    List all available text-to-speech voices with metadata. Returns: dict with keys: - voices (list): Available voices, each with id, name, gender, accent, grade - defaultVoice (str): Default voice ID

    Annotations the server declares

    • title: List TTS Voices
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true

This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.

Install

Pinned to version 2.3.0. Values in angle brackets are yours to fill in.

claude mcp add --transport http speech-ai https://apim-ai-apis.azure-api.net/mcp/pronunciation/mcp

Run in a terminal.

Claude Code permissions

Allow the reads, ask before a change, deny the deletes. Paste into .claude/settings.json.

.claude/settings.json
{
  "permissions": {
    "allow": [
      "mcp__speech-ai__get_phoneme_inventory",
      "mcp__speech-ai__list_tts_voices"
    ],
    "ask": [
      "mcp__speech-ai__assess_pronunciation",
      "mcp__speech-ai__check_pronunciation_service",
      "mcp__speech-ai__transcribe_audio",
      "mcp__speech-ai__check_stt_service",
      "mcp__speech-ai__synthesize_speech",
      "mcp__speech-ai__check_tts_service",
      "mcp__speech-ai__transcribe_audio_pro",
      "mcp__speech-ai__check_whisper_service"
    ],
    "deny": []
  }
}

Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. It assumes the server is installed as speech-ai. Review it before you use it.