solace-agent-mesh 0.2.4__py3-none-any.whl → 1.0.1__py3-none-any.whl

solace_agent_mesh/agent/tools/audio_tools.py

@@ -0,0 +1,1622 @@
+"""
+Collection of Python tools for audio processing and text-to-speech generation.
+"""
+
+import asyncio
+import inspect
+import io
+import json
+import os
+import random
+import tempfile
+import uuid
+import wave
+from datetime import datetime, timezone
+from typing import Any, Dict, List, Optional
+
+import httpx
+
+from google import genai
+from google.genai import types as adk_types
+from google.adk.tools import ToolContext
+from pydub import AudioSegment
+from solace_ai_connector.common.log import log
+
+from ...agent.utils.artifact_helpers import (
+    load_artifact_content_or_metadata,
+    save_artifact_with_metadata,
+    DEFAULT_SCHEMA_MAX_KEYS,
+)
+from ...agent.utils.context_helpers import get_original_session_id
+
+from .tool_definition import BuiltinTool
+from .registry import tool_registry
+
+
+VOICE_TONE_MAPPING = {
+    "bright": ["Zephyr", "Autonoe"],
+    "upbeat": ["Puck", "Laomedeia"],
+    "informative": ["Charon", "Rasalgethi"],
+    "firm": ["Kore", "Orus", "Alnilam"],
+    "excitable": ["Fenrir"],
+    "youthful": ["Leda"],
+    "breezy": ["Aoede"],
+    "easy-going": ["Callirhoe", "Umbriel"],
+    "breathy": ["Enceladus"],
+    "clear": ["Iapetus", "Erinome"],
+    "smooth": ["Algieba", "Despina"],
+    "gravelly": ["Algenib"],
+    "soft": ["Achernar"],
+    "even": ["Schedar"],
+    "mature": ["Gacrux"],
+    "forward": ["Pulcherrima"],
+    "friendly": ["Achird"],
+    "casual": ["Zubenelgenubi"],
+    "gentle": ["Vindemiatrix"],
+    "lively": ["Sadachbia"],
+    "knowledgeable": ["Sadaltager"],
+    "warm": ["Sulafat"],
+}
+
+GENDER_TO_VOICE_MAPPING = {
+    "male": [
+        "Puck",
+        "Charon",
+        "Orus",
+        "Enceladus",
+        "Iapetus",
+        "Algieba",
+        "Algenib",
+        "Alnilam",
+        "Schedar",
+        "Achird",
+        "Zubenelgenubi",
+        "Sadachbia",
+        "Sadaltager",
+    ],
+    "female": [
+        "Zephyr",
+        "Kore",
+        "Leda",
+        "Aoede",
+        "Callirhoe",
+        "Autonoe",
+        "Despina",
+        "Erinome",
+        "Laomedeia",
+        "Achernar",
+        "Gacrux",
+        "Pulcherrima",
+        "Vindemiatrix",
+        "Sulafat",
+    ],
+    "neutral": [
+        "Fenrir",
+        "Umbriel",
+        "Rasalgethi",
+    ],
+}
+
+_all_voices_set = set(
+    voice for voices_in_tone in VOICE_TONE_MAPPING.values() for voice in voices_in_tone
+)
+_all_voices_set.update(
+    voice
+    for voices_in_gender in GENDER_TO_VOICE_MAPPING.values()
+    for voice in voices_in_gender
+)
+ALL_AVAILABLE_VOICES = list(_all_voices_set)
+
+if not ALL_AVAILABLE_VOICES:
+    ALL_AVAILABLE_VOICES = [
+        "Kore",
+        "Puck",
+        "Zephyr",
+        "Charon",
+        "Rasalgethi",
+        "Alnilam",
+        "Fenrir",
+        "Leda",
+        "Aoede",
+        "Callirhoe",
+        "Umbriel",
+        "Enceladus",
+        "Iapetus",
+        "Erinome",
+        "Algieba",
+        "Despina",
+        "Algenib",
+        "Achernar",
+        "Schedar",
+        "Gacrux",
+        "Pulcherrima",
+        "Achird",
+        "Zubenelgenubi",
+        "Vindemiatrix",
+        "Sadachbia",
+        "Sadaltager",
+        "Sulafat",
+        "Autonoe",
+        "Laomedeia",
+        "Orus",
+    ]
+
+
+SUPPORTED_LANGUAGES = {
+    "arabic": "ar-EG",
+    "arabic_egyptian": "ar-EG",
+    "german": "de-DE",
+    "english": "en-US",
+    "english_us": "en-US",
+    "english_india": "en-IN",
+    "spanish": "es-US",
+    "spanish_us": "es-US",
+    "french": "fr-FR",
+    "hindi": "hi-IN",
+    "indonesian": "id-ID",
+    "italian": "it-IT",
+    "japanese": "ja-JP",
+    "korean": "ko-KR",
+    "portuguese": "pt-BR",
+    "portuguese_brazil": "pt-BR",
+    "russian": "ru-RU",
+    "dutch": "nl-NL",
+    "polish": "pl-PL",
+    "thai": "th-TH",
+    "turkish": "tr-TR",
+    "vietnamese": "vi-VN",
+    "romanian": "ro-RO",
+    "ukrainian": "uk-UA",
+    "bengali": "bn-BD",
+    "marathi": "mr-IN",
+    "tamil": "ta-IN",
+    "telugu": "te-IN",
+}
+
+from typing import Set
+
+DEFAULT_VOICE = "Kore"
+
+
+def _get_effective_tone_voices(
+    tone: Optional[str], voice_tone_mapping: Optional[Dict[str, List[str]]] = None
+) -> Optional[List[str]]:
+    """Helper to get voices for a tone, considering aliases."""
+    if not tone:
+        return None
+    mapping = voice_tone_mapping or VOICE_TONE_MAPPING
+    tone_lower = tone.lower().strip()
+    tone_aliases = {
+        "professional": "firm",
+        "business": "firm",
+        "corporate": "firm",
+        "cheerful": "upbeat",
+        "happy": "upbeat",
+        "energetic": "excitable",
+        "calm": "soft",
+        "relaxed": "easy-going",
+        "serious": "informative",
+        "educational": "knowledgeable",
+        "teaching": "knowledgeable",
+        "conversational": "casual",
+        "natural": "friendly",
+        "welcoming": "warm",
+    }
+    effective_tone = tone_aliases.get(tone_lower, tone_lower)
+    return mapping.get(effective_tone)
+
+
+def _get_gender_voices(
+    gender: Optional[str], gender_voice_mapping: Optional[Dict[str, List[str]]] = None
+) -> Optional[List[str]]:
+    """Helper to get voices for a gender."""
+    if not gender:
+        return None
+    mapping = gender_voice_mapping or GENDER_TO_VOICE_MAPPING
+    return mapping.get(gender.lower().strip())
+
+
+def _get_voice_for_speaker(
+    gender: Optional[str],
+    tone: Optional[str],
+    used_voices_in_current_call: Set[str],
+    voice_tone_mapping: Optional[Dict[str, List[str]]] = None,
+    gender_voice_mapping: Optional[Dict[str, List[str]]] = None,
+) -> str:
+    """
+    Selects a voice based on desired gender and/or tone, prioritizing uniqueness.
+
+    Selection Priority:
+    1. Unique voice matching both specified gender and tone.
+    2. Unique voice matching specified gender (if tone match failed or tone not specified).
+    3. Unique voice matching specified tone (if gender match failed or gender not specified).
+    4. Any unique voice from the global pool.
+    5. If all unique options exhausted (reuse necessary):
+        a. Voice matching both specified gender and tone.
+        b. Voice matching specified gender.
+        c. Voice matching specified tone.
+        d. Any voice from the global pool.
+    6. Fallback to DEFAULT_VOICE ("Kore").
+
+    Args:
+        gender: The desired gender ("male", "female", "neutral").
+        tone: The desired tone (e.g., "friendly", "professional").
+        used_voices_in_current_call: A set of voice names already used.
+        voice_tone_mapping: Optional custom tone-to-voice mapping.
+        gender_voice_mapping: Optional custom gender-to-voice mapping.
+
+    Returns:
+        A voice name string.
+    """
+    log_identifier_select = "[AudioTools:_get_voice_for_speaker]"
+    log.debug(
+        "%s Selecting voice for gender='%s', tone='%s', used_voices=%s",
+        log_identifier_select,
+        gender,
+        tone,
+        used_voices_in_current_call,
+    )
+
+    candidate_pool = list(ALL_AVAILABLE_VOICES)
+
+    gender_specific_voices = None
+    if gender:
+        gender_specific_voices = _get_gender_voices(gender, gender_voice_mapping)
+        if gender_specific_voices:
+            candidate_pool = [v for v in candidate_pool if v in gender_specific_voices]
+            log.debug(
+                "%s Filtered by gender '%s'. Candidates: %s",
+                log_identifier_select,
+                gender,
+                candidate_pool,
+            )
+        else:
+            log.warning(
+                "%s Gender '%s' not found in mapping or has no voices. Gender filter ignored for now.",
+                log_identifier_select,
+                gender,
+            )
+            gender_specific_voices = None
+
+    tone_specific_voices = None
+    if tone:
+        voices_for_tone = _get_effective_tone_voices(tone, voice_tone_mapping)
+        if voices_for_tone:
+            tone_specific_voices = voices_for_tone
+            candidate_pool = [v for v in candidate_pool if v in voices_for_tone]
+            log.debug(
+                "%s Filtered by tone '%s'. Candidates: %s",
+                log_identifier_select,
+                tone,
+                candidate_pool,
+            )
+        else:
+            log.warning(
+                "%s Tone '%s' not found in mapping or has no voices. Tone filter ignored.",
+                log_identifier_select,
+                tone,
+            )
+            tone_specific_voices = None
+
+    available_unique_voices = [
+        v for v in candidate_pool if v not in used_voices_in_current_call
+    ]
+    if available_unique_voices:
+        selected = random.choice(available_unique_voices)
+        log.info(
+            "%s Selected unique voice '%s' from gender/tone filtered pool.",
+            log_identifier_select,
+            selected,
+        )
+        return selected
+
+    log.debug(
+        "%s No unique voice in primary filtered pool. Trying fallbacks for uniqueness.",
+        log_identifier_select,
+    )
+    if gender_specific_voices:
+        available_gender_unique = [
+            v for v in gender_specific_voices if v not in used_voices_in_current_call
+        ]
+        if available_gender_unique:
+            selected = random.choice(available_gender_unique)
+            log.info(
+                "%s Selected unique voice '%s' from gender-only pool (tone constraint relaxed).",
+                log_identifier_select,
+                selected,
+            )
+            return selected
+    if tone_specific_voices:
+        available_tone_unique = [
+            v for v in tone_specific_voices if v not in used_voices_in_current_call
+        ]
+        if available_tone_unique:
+            selected = random.choice(available_tone_unique)
+            log.info(
+                "%s Selected unique voice '%s' from tone-only pool (gender constraint relaxed or not specified).",
+                log_identifier_select,
+                selected,
+            )
+            return selected
+    globally_available_unique = [
+        v for v in ALL_AVAILABLE_VOICES if v not in used_voices_in_current_call
+    ]
+    if globally_available_unique:
+        selected = random.choice(globally_available_unique)
+        log.info(
+            "%s Selected unique voice '%s' from global pool (all constraints relaxed).",
+            log_identifier_select,
+            selected,
+        )
+        return selected
+
+    log.warning("%s All voices are used. Reusing a voice.", log_identifier_select)
+    if candidate_pool:
+        selected = random.choice(candidate_pool)
+        log.info(
+            "%s Reusing voice '%s' from gender/tone filtered pool.",
+            log_identifier_select,
+            selected,
+        )
+        return selected
+    if gender_specific_voices:
+        selected = random.choice(gender_specific_voices)
+        log.info(
+            "%s Reusing voice '%s' from gender-only pool.",
+            log_identifier_select,
+            selected,
+        )
+        return selected
+    if tone_specific_voices:
+        selected = random.choice(tone_specific_voices)
+        log.info(
+            "%s Reusing voice '%s' from tone-only pool.",
+            log_identifier_select,
+            selected,
+        )
+        return selected
+    if ALL_AVAILABLE_VOICES:
+        selected = random.choice(ALL_AVAILABLE_VOICES)
+        log.info(
+            "%s Reusing voice '%s' from global pool.", log_identifier_select, selected
+        )
+        return selected
+
+    log.error(
+        "%s No voices available in any mapping or pool. Using default '%s'.",
+        log_identifier_select,
+        DEFAULT_VOICE,
+    )
+    return DEFAULT_VOICE
+
+
+def _get_language_code(language: str) -> str:
+    """
+    Get BCP-47 language code from language name or code.
+
+    Args:
+        language: Language name or BCP-47 code
+
+    Returns:
+        BCP-47 language code, defaults to "en-US"
+    """
+    if not language:
+        return "en-US"
+
+    language_lower = language.lower().strip()
+
+    if "-" in language_lower and len(language_lower) >= 5:
+        return language
+
+    if language_lower in SUPPORTED_LANGUAGES:
+        return SUPPORTED_LANGUAGES[language_lower]
+
+    log.warning(f"[AudioTools] Unknown language '{language}', using default 'en-US'")
+    return "en-US"
+
+
+def _create_voice_config(voice_name: str) -> adk_types.VoiceConfig:
+    """Create a VoiceConfig for single-voice TTS."""
+    return adk_types.VoiceConfig(
+        prebuilt_voice_config=adk_types.PrebuiltVoiceConfig(voice_name=voice_name)
+    )
+
+
+def _create_multi_speaker_config(
+    speaker_configs: List[Dict[str, str]],
+) -> adk_types.MultiSpeakerVoiceConfig:
+    """Create a MultiSpeakerVoiceConfig for multi-speaker TTS."""
+    speaker_voice_configs = []
+
+    for config in speaker_configs:
+        speaker_name = config.get("name", "Speaker")
+        voice_name = config.get("voice", "Kore")
+
+        speaker_voice_config = adk_types.SpeakerVoiceConfig(
+            speaker=speaker_name, voice_config=_create_voice_config(voice_name)
+        )
+        speaker_voice_configs.append(speaker_voice_config)
+
+    return adk_types.MultiSpeakerVoiceConfig(
+        speaker_voice_configs=speaker_voice_configs
+    )
+
+
+async def _generate_audio_with_gemini(
+    client: genai.Client,
+    prompt: str,
+    speech_config: adk_types.SpeechConfig,
+    model: str = "gemini-2.5-flash-preview-tts",
+    language: str = "en-US",
+) -> bytes:
+    """
+    Shared function for generating audio using Gemini API.
+
+    Args:
+        client: Gemini client instance
+        prompt: Text prompt for TTS
+        speech_config: Speech configuration (single or multi-speaker)
+        model: Gemini model to use
+        language: BCP-47 language code
+
+    Returns:
+        Raw audio data as bytes
+    """
+    config = adk_types.GenerateContentConfig(
+        response_modalities=["AUDIO"],
+        speech_config=speech_config,
+    )
+
+    if hasattr(config, "language"):
+        config.language = language
+
+    response = await asyncio.to_thread(
+        client.models.generate_content, model=model, contents=prompt, config=config
+    )
+
+    if (
+        not response
+        or not response.candidates
+        or not response.candidates[0].content.parts
+    ):
+        raise ValueError("Gemini API did not return valid audio data")
+
+    audio_data = response.candidates[0].content.parts[0].inline_data.data
+    if not audio_data:
+        raise ValueError("No audio data received from Gemini API")
+
+    return audio_data
+
+
+def _create_wav_file(
+    filename: str,
+    pcm_data: bytes,
+    channels: int = 1,
+    rate: int = 24000,
+    sample_width: int = 2,
+):
+    """Create a proper WAV file from PCM data (based on working example)."""
+    with wave.open(filename, "wb") as wf:
+        wf.setnchannels(channels)
+        wf.setsampwidth(sample_width)
+        wf.setframerate(rate)
+        wf.writeframes(pcm_data)
+
+
+async def _convert_pcm_to_mp3(pcm_data: bytes) -> bytes:
+    """
+    Shared function for converting raw PCM data to MP3 format.
+
+    Args:
+        pcm_data: Raw PCM audio data from Gemini API
+
+    Returns:
+        MP3 audio data as bytes
+    """
+    wav_temp_path = None
+    mp3_temp_path = None
+
+    try:
+        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as wav_temp:
+            wav_temp_path = wav_temp.name
+
+        await asyncio.to_thread(_create_wav_file, wav_temp_path, pcm_data)
+
+        with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as mp3_temp:
+            mp3_temp_path = mp3_temp.name
+
+        audio = await asyncio.to_thread(AudioSegment.from_wav, wav_temp_path)
+        await asyncio.to_thread(audio.export, mp3_temp_path, format="mp3")
+
+        with open(mp3_temp_path, "rb") as mp3_file:
+            mp3_data = mp3_file.read()
+
+        return mp3_data
+
+    finally:
+        if wav_temp_path:
+            try:
+                os.remove(wav_temp_path)
+            except OSError:
+                pass
+        if mp3_temp_path:
+            try:
+                os.remove(mp3_temp_path)
+            except OSError:
+                pass
+
+
+async def _save_audio_artifact(
+    audio_data: bytes,
+    filename: str,
+    metadata: Dict[str, Any],
+    tool_context: ToolContext,
+) -> Dict[str, Any]:
+    """
+    Shared function for saving audio artifacts with metadata.
+
+    Args:
+        audio_data: Audio data to save
+        filename: Filename for the artifact
+        metadata: Metadata dictionary
+        tool_context: ADK tool context
+
+    Returns:
+        Save result dictionary
+    """
+    inv_context = tool_context._invocation_context
+    app_name = inv_context.app_name
+    user_id = inv_context.user_id
+    session_id = get_original_session_id(inv_context)
+    artifact_service = inv_context.artifact_service
+
+    if not artifact_service:
+        raise ValueError("ArtifactService is not available in the context")
+
+    save_result = await save_artifact_with_metadata(
+        artifact_service=artifact_service,
+        app_name=app_name,
+        user_id=user_id,
+        session_id=session_id,
+        filename=filename,
+        content_bytes=audio_data,
+        mime_type="audio/mpeg",
+        metadata_dict=metadata,
+        timestamp=datetime.now(timezone.utc),
+        schema_max_keys=DEFAULT_SCHEMA_MAX_KEYS,
+        tool_context=tool_context,
+    )
+
+    if save_result.get("status") == "error":
+        raise IOError(
+            f"Failed to save audio artifact: {save_result.get('message', 'Unknown error')}"
+        )
+
+    return save_result
+
+
+async def select_voice(
+    gender: Optional[str] = None,
+    tone: Optional[str] = None,
+    exclude_voices: Optional[List[str]] = None,
+    tool_context: ToolContext = None,
+    tool_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """
+    Selects a suitable voice name based on criteria like gender and tone.
+    Use this to get a consistent voice name that can be passed to the `text_to_speech` tool for multiple calls.
+
+    Args:
+        gender: Optional desired gender for the voice ("male", "female", "neutral").
+        tone: Optional tone preference (e.g., "friendly", "professional", "warm").
+        exclude_voices: Optional list of voice names to exclude from the selection.
+        tool_context: ADK tool context.
+        tool_config: Configuration including voice mappings.
+
+    Returns:
+        Dictionary with status and the selected voice name.
+    """
+    log_identifier = "[AudioTools:select_voice]"
+    log.info(
+        f"{log_identifier} Selecting voice for gender='{gender}', tone='{tone}', excluding='{exclude_voices}'"
+    )
+
+    try:
+        config = tool_config or {}
+        voice_tone_mapping = config.get("voice_tone_mapping", VOICE_TONE_MAPPING)
+        gender_voice_mapping = config.get(
+            "gender_voice_mapping", GENDER_TO_VOICE_MAPPING
+        )
+
+        used_voices = set(exclude_voices) if exclude_voices else set()
+
+        selected_voice = _get_voice_for_speaker(
+            gender=gender,
+            tone=tone,
+            used_voices_in_current_call=used_voices,
+            voice_tone_mapping=voice_tone_mapping,
+            gender_voice_mapping=gender_voice_mapping,
+        )
+
+        log.info(f"{log_identifier} Selected voice: {selected_voice}")
+
+        return {
+            "status": "success",
+            "message": f"Successfully selected voice '{selected_voice}'.",
+            "voice_name": selected_voice,
+        }
+
+    except Exception as e:
+        log.exception(f"{log_identifier} Unexpected error in select_voice: {e}")
+        return {"status": "error", "message": f"An unexpected error occurred: {e}"}
+
+
+async def text_to_speech(
+    text: str,
+    output_filename: Optional[str] = None,
+    voice_name: Optional[str] = None,
+    gender: Optional[str] = None,
+    tone: Optional[str] = None,
+    language: Optional[str] = None,
+    tool_context: ToolContext = None,
+    tool_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """
+    Converts text to speech using Gemini TTS API and saves as MP3 artifact.
+
+    Args:
+        text: The text to convert to speech.
+        output_filename: Optional filename for the output MP3.
+        voice_name: Optional specific voice name (e.g., "Kore", "Puck"). Overrides gender and tone.
+        gender: Optional desired gender for the voice ("male", "female", "neutral").
+                Used if 'voice_name' is not provided.
+        tone: Optional tone preference (e.g., "friendly", "professional", "warm").
+              Used if 'voice_name' is not provided, considered after 'gender'.
+        language: Optional language code (e.g., "en-US", "fr-FR", "ja-JP").
+        tool_context: ADK tool context.
+        tool_config: Configuration including API key, model settings, and voice mappings.
+
+    Returns:
+        Dictionary with status, output filename, version, and preview.
+    """
+    log_identifier = "[AudioTools:text_to_speech]"
+
+    if not tool_context:
+        return {"status": "error", "message": "ToolContext is missing"}
+
+    if not text or not text.strip():
+        return {"status": "error", "message": "Text input is required"}
+
+    try:
+        log.info(f"{log_identifier} Processing TTS request for text: '{text[:50]}...'")
+
+        config = tool_config or {}
+        api_key = config.get("gemini_api_key")
+        model = config.get("model", "gemini-2.5-flash-preview-tts")
+        default_voice = config.get("voice_name", DEFAULT_VOICE)
+        default_language = config.get("language", "en-US")
+        voice_tone_mapping = config.get("voice_tone_mapping", VOICE_TONE_MAPPING)
+        gender_voice_mapping = config.get(
+            "gender_voice_mapping", GENDER_TO_VOICE_MAPPING
+        )
+
+        if not api_key:
+            return {
+                "status": "error",
+                "message": "GEMINI_API_KEY is required in tool configuration",
+            }
+
+        final_voice = voice_name
+        if not final_voice:
+            final_voice = _get_voice_for_speaker(
+                gender=gender,
+                tone=tone,
+                used_voices_in_current_call=set(),
+                voice_tone_mapping=voice_tone_mapping,
+                gender_voice_mapping=gender_voice_mapping,
+            )
+            log.info(
+                f"{log_identifier} Selected voice '{final_voice}' for gender='{gender}', tone='{tone}'"
+            )
+        else:
+            log.info(
+                f"{log_identifier} Using specified voice_name '{final_voice}'. Gender/tone selection skipped."
+            )
+
+        if not final_voice:
+            final_voice = default_voice
+            log.warning(
+                f"{log_identifier} Voice selection resulted in None, using default '{default_voice}'."
+            )
+
+        final_language = _get_language_code(language or default_language)
+
+        client = genai.Client(api_key=api_key)
+
+        voice_config = _create_voice_config(final_voice)
+        speech_config = adk_types.SpeechConfig(voice_config=voice_config)
+
+        log.info(
+            f"{log_identifier} Generating audio with voice '{final_voice}' and language '{final_language}'"
+        )
+        wav_data = await _generate_audio_with_gemini(
+            client=client,
+            prompt=f"Say in a clear voice: {text}",
+            speech_config=speech_config,
+            model=model,
+            language=final_language,
+        )
+
+        log.info(f"{log_identifier} Converting audio to MP3 format")
+        mp3_data = await _convert_pcm_to_mp3(wav_data)
+
+        final_filename = output_filename
+        if not final_filename:
+            final_filename = f"tts_audio_{uuid.uuid4()}.mp3"
+        elif not final_filename.lower().endswith(".mp3"):
+            final_filename = f"{final_filename}.mp3"
+
+        metadata = {
+            "description": f"Text-to-speech audio generated from text input. Used voice: {final_voice}\nSource text: {text[:1500]}",
+            "source_text": text[:500],
+            "voice_name": final_voice,
+            "language": final_language,
+            "model": model,
+            "generation_tool": "text_to_speech",
+            "generation_timestamp": datetime.now(timezone.utc).isoformat(),
+        }
+
+        if voice_name:
+            metadata["requested_voice_name"] = voice_name
+        if gender:
+            metadata["requested_gender"] = gender
+        if tone:
+            metadata["requested_tone"] = tone
+        if language:
+            metadata["requested_language"] = language
+
+        log.info(f"{log_identifier} Saving audio artifact '{final_filename}'")
+        save_result = await _save_audio_artifact(
+            audio_data=mp3_data,
+            filename=final_filename,
+            metadata=metadata,
+            tool_context=tool_context,
+        )
+
+        log.info(
+            f"{log_identifier} Audio artifact saved successfully: {final_filename} v{save_result['data_version']}"
+        )
+
+        return {
+            "status": "success",
+            "message": f"Text-to-speech audio generated and saved successfully",
+            "output_filename": final_filename,
+            "output_version": save_result["data_version"],
+            "voice_used": final_voice,
+            "language_used": final_language,
+            "result_preview": f"Audio '{final_filename}' (v{save_result['data_version']}) created from text: \"{text[:100]}...\"",
+        }
+
+    except ValueError as ve:
+        log.error(f"{log_identifier} Value error: {ve}")
+        return {"status": "error", "message": str(ve)}
+    except IOError as ioe:
+        log.error(f"{log_identifier} IO error: {ioe}")
+        return {"status": "error", "message": str(ioe)}
+    except Exception as e:
+        log.exception(f"{log_identifier} Unexpected error in text_to_speech: {e}")
+        return {"status": "error", "message": f"An unexpected error occurred: {e}"}
+
+
+async def multi_speaker_text_to_speech(
+    conversation_text: str,
+    output_filename: Optional[str] = None,
+    speaker_configs: Optional[List[Dict[str, str]]] = None,
+    language: Optional[str] = None,
+    tool_context: ToolContext = None,
+    tool_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """
+    Converts conversation text with speaker labels to speech using multiple voices.
+
+    Args:
+        conversation_text: Text with speaker labels (e.g., "Speaker1: Hello\\nSpeaker2: Hi there").
+        output_filename: Optional filename for the output MP3.
+        speaker_configs: Optional list of speaker configurations. Each item is a dictionary:
+                         `{"name": "Speaker1", "voice": "Kore", "gender": "female", "tone": "firm"}`.
+                         - `name` (str): The speaker's name as it appears in `conversation_text`.
+                         - `voice` (Optional[str]): Specific voice name to use. Overrides gender/tone.
+                         - `gender` (Optional[str]): Desired gender ("male", "female", "neutral").
+                         - `tone` (Optional[str]): Desired tone (e.g., "friendly").
+                         If only `gender` and/or `tone` are provided, a voice is selected.
+                         If no config for a speaker in text, or if speaker_configs is empty,
+                         default speakers from tool_config are used, or a default voice is assigned.
+        language: Optional language code (e.g., "en-US", "fr-FR", "ja-JP").
+        tool_context: ADK tool context.
+        tool_config: Configuration including API key, model, default speakers, and voice mappings.
+
+    Returns:
+        Dictionary with status, output filename, version, and preview.
+    """
+    log_identifier = "[AudioTools:multi_speaker_text_to_speech]"
+
+    if not tool_context:
+        return {"status": "error", "message": "ToolContext is missing"}
+
+    if not conversation_text or not conversation_text.strip():
+        return {"status": "error", "message": "Conversation text input is required"}
+
+    try:
+        log.info(
+            f"{log_identifier} Processing multi-speaker TTS request for text: '{conversation_text[:50]}...'"
+        )
+
+        config = tool_config or {}
+        api_key = config.get("gemini_api_key")
+        model = config.get("model", "gemini-2.5-flash-preview-tts")
+        default_language = config.get("language", "en-US")
+        default_speakers = config.get(
+            "default_speakers",
+            [
+                {"name": "Speaker1", "voice": "Kore"},
+                {
+                    "name": "Speaker2",
+                    "voice": "Puck",
+                    "gender": "male",
+                    "tone": "upbeat",
+                },
+                {
+                    "name": "Speaker3",
+                    "voice": "Zephyr",
+                    "gender": "female",
+                    "tone": "bright",
+                },
+            ],
+        )
+        voice_tone_mapping = config.get("voice_tone_mapping", VOICE_TONE_MAPPING)
+        gender_voice_mapping = config.get(
+            "gender_voice_mapping", GENDER_TO_VOICE_MAPPING
+        )
+
+        if not api_key:
+            return {
+                "status": "error",
+                "message": "GEMINI_API_KEY is required in tool configuration",
+            }
+
+        final_speaker_configs = []
+        used_voices_in_current_call: Set[str] = set()
+
+        configs_to_process = speaker_configs if speaker_configs else default_speakers
+        if not configs_to_process and conversation_text:
+            log.warning(
+                "%s No speaker_configs and no default_speakers. Creating a default speaker.",
+                log_identifier,
+            )
+            configs_to_process = [
+                {"name": "Speaker1", "gender": "neutral", "tone": "neutral"}
+            ]
+
+        for i, config_item in enumerate(configs_to_process):
+            speaker_name = config_item.get("name", f"Speaker{i+1}")
+            voice_name_from_config = config_item.get("voice")
+            gender_from_config = config_item.get("gender")
+            tone_from_config = config_item.get("tone")
+            final_voice_for_speaker = None
+
+            if voice_name_from_config:
+                final_voice_for_speaker = voice_name_from_config
+                log.info(
+                    f"{log_identifier} Using specified voice '{final_voice_for_speaker}' for speaker '{speaker_name}'."
+                )
+            else:
+                final_voice_for_speaker = _get_voice_for_speaker(
+                    gender=gender_from_config,
+                    tone=tone_from_config,
+                    used_voices_in_current_call=used_voices_in_current_call,
+                    voice_tone_mapping=voice_tone_mapping,
+                    gender_voice_mapping=gender_voice_mapping,
+                )
+                log.info(
+                    f"{log_identifier} Selected voice '{final_voice_for_speaker}' for speaker '{speaker_name}' (gender='{gender_from_config}', tone='{tone_from_config}')."
+                )
+
+            if not final_voice_for_speaker:
+                final_voice_for_speaker = DEFAULT_VOICE
+                log.warning(
+                    f"{log_identifier} Voice selection for speaker '{speaker_name}' resulted in None, using default '{DEFAULT_VOICE}'."
+                )
+
+            final_speaker_configs.append(
+                {"name": speaker_name, "voice": final_voice_for_speaker}
+            )
+            if final_voice_for_speaker:
+                used_voices_in_current_call.add(final_voice_for_speaker)
+
+        final_language = _get_language_code(language or default_language)
+
+        client = genai.Client(api_key=api_key)
+
+        multi_speaker_config = _create_multi_speaker_config(final_speaker_configs)
+        speech_config = adk_types.SpeechConfig(
+            multi_speaker_voice_config=multi_speaker_config
+        )
+
+        log.info(
+            f"{log_identifier} Generating multi-speaker audio with {len(final_speaker_configs)} speakers and language '{final_language}'"
+        )
+        wav_data = await _generate_audio_with_gemini(
+            client=client,
+            prompt=f"TTS the following conversation: {conversation_text}",
+            speech_config=speech_config,
+            model=model,
+            language=final_language,
+        )
+
+        log.info(f"{log_identifier} Converting audio to MP3 format")
+        mp3_data = await _convert_pcm_to_mp3(wav_data)
+
+        final_filename = output_filename
+        if not final_filename:
+            final_filename = f"multi_speaker_tts_{uuid.uuid4()}.mp3"
+        elif not final_filename.lower().endswith(".mp3"):
+            final_filename = f"{final_filename}.mp3"
+
+        voice_info = ", ".join(
+            [f"{s['name']} ({s['voice']})" for s in final_speaker_configs]
+        )
+
+        metadata = {
+            "description": f"Multi-speaker text-to-speech audio generated from conversation. Voices used: {voice_info}\nSource text: {conversation_text[:1500]}...",
+            "source_text": conversation_text[:500],
+            "language": final_language,
+            "model": model,
+            "generation_tool": "multi_speaker_text_to_speech",
+            "generation_timestamp": datetime.now(timezone.utc).isoformat(),
+            "speaker_count": len(final_speaker_configs),
+            "speakers_used_details": json.dumps(final_speaker_configs),
+        }
+
+        if language:
+            metadata["requested_language"] = language
+        if speaker_configs:
+            metadata["requested_speaker_configs"] = json.dumps(speaker_configs)
+
+        log.info(f"{log_identifier} Saving audio artifact '{final_filename}'")
+        save_result = await _save_audio_artifact(
+            audio_data=mp3_data,
+            filename=final_filename,
+            metadata=metadata,
+            tool_context=tool_context,
+        )
+
+        log.info(
+            f"{log_identifier} Multi-speaker audio artifact saved successfully: {final_filename} v{save_result['data_version']}"
+        )
+
+        speaker_summary = ", ".join(
+            [f"{s['name']} ({s['voice']})" for s in final_speaker_configs]
+        )
+
+        return {
+            "status": "success",
+            "message": f"Multi-speaker text-to-speech audio generated and saved successfully: Speakers: {speaker_summary}",
+            "output_filename": final_filename,
+            "output_version": save_result["data_version"],
+            "speakers_used": speaker_summary,
+            "language_used": final_language,
+            "result_preview": f"Multi-speaker audio '{final_filename}' (v{save_result['data_version']}) created with {len(final_speaker_configs)} speakers from conversation: \"{conversation_text[:100]}...\"",
+        }
+
+    except ValueError as ve:
+        log.error(f"{log_identifier} Value error: {ve}")
+        return {"status": "error", "message": str(ve)}
+    except IOError as ioe:
+        log.error(f"{log_identifier} IO error: {ioe}")
+        return {"status": "error", "message": str(ioe)}
+    except Exception as e:
+        log.exception(
+            f"{log_identifier} Unexpected error in multi_speaker_text_to_speech: {e}"
+        )
+        return {"status": "error", "message": f"An unexpected error occurred: {e}"}
+
+
+def _is_supported_audio_format_for_transcription(filename: str) -> bool:
+    """Check if the audio format is supported for transcription."""
+    ext = os.path.splitext(filename)[1].lower()
+    supported_formats = {".wav", ".mp3"}
+    return ext in supported_formats
+
+
+def _get_audio_mime_type(filename: str) -> str:
+    """Get MIME type from audio file extension."""
+    ext = os.path.splitext(filename)[1].lower()
+    mime_mapping = {".wav": "audio/wav", ".mp3": "audio/mpeg"}
+    return mime_mapping.get(ext, "audio/wav")
+
+
+async def concatenate_audio(
+    clips_to_join: List[Dict[str, Any]],
+    output_filename: Optional[str] = None,
+    tool_context: ToolContext = None,
+    tool_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """
+    Combines multiple audio artifacts in a specified order into a single audio file.
+    Allows for custom pause durations between each clip.
+
+    Args:
+        clips_to_join: An ordered list of clip objects to be joined. Each object should contain:
+                       - `filename` (str): The artifact filename of the audio clip (with optional :version).
+                       - `pause_after_ms` (Optional[int]): The duration of silence, in milliseconds,
+                         to insert *after* this clip. If omitted, a default pause will be used. The gap between
+                         two people speaking in a conversation is typically around 250ms.
+        output_filename: Optional. The desired filename for the final combined audio artifact.
+        tool_context: The context provided by the ADK framework.
+        tool_config: Configuration dictionary.
+
+    Returns:
+        A dictionary with status and output artifact details.
+    """
+    log_identifier = "[AudioTools:concatenate_audio]"
+    if not tool_context:
+        return {"status": "error", "message": "ToolContext is missing"}
+    if not clips_to_join:
+        return {
+            "status": "error",
+            "message": "The 'clips_to_join' list cannot be empty.",
+        }
+
+    try:
+        inv_context = tool_context._invocation_context
+        app_name = inv_context.app_name
+        user_id = inv_context.user_id
+        session_id = get_original_session_id(inv_context)
+        artifact_service = inv_context.artifact_service
+        host_component = getattr(inv_context.agent, "host_component", None)
+
+        if not artifact_service:
+            raise ValueError("ArtifactService is not available in the context.")
+
+        config = tool_config or {}
+        default_pause_ms = config.get("default_pause_ms", 250)
+
+        combined_audio = None
+        source_filenames = []
+
+        for i, clip_info in enumerate(clips_to_join):
+            clip_filename_with_version = clip_info.get("filename")
+            if not clip_filename_with_version:
+                raise ValueError(
+                    f"Clip at index {i} is missing the required 'filename' key."
+                )
+
+            source_filenames.append(clip_filename_with_version)
+
+            parts = clip_filename_with_version.rsplit(":", 1)
+            filename_base = parts[0]
+            version_str = None
+            if len(parts) > 1 and parts[1].isdigit():
+                version_str = parts[1]
+            version_to_load = int(version_str) if version_str else "latest"
+
+            load_result = await load_artifact_content_or_metadata(
+                artifact_service=artifact_service,
+                app_name=app_name,
+                user_id=user_id,
+                session_id=session_id,
+                filename=filename_base,
+                version=version_to_load,
+                return_raw_bytes=True,
+                component=host_component,
+                log_identifier_prefix=f"{log_identifier}[LoadClip:{clip_filename_with_version}]",
+            )
+
+            if load_result.get("status") != "success":
+                raise FileNotFoundError(
+                    f"Failed to load audio clip '{clip_filename_with_version}': {load_result.get('message')}"
+                )
+
+            audio_bytes = load_result.get("raw_bytes")
+            mime_type = load_result.get("mime_type", "audio/mpeg")
+
+            audio_format = "mp3"
+            if "wav" in mime_type:
+                audio_format = "wav"
+            elif "mpeg" in mime_type:
+                audio_format = "mp3"
+
+            log.debug(
+                f"{log_identifier} Loading clip '{clip_filename_with_version}' with format '{audio_format}'"
+            )
+
+            clip_segment = await asyncio.to_thread(
+                AudioSegment.from_file, io.BytesIO(audio_bytes), format=audio_format
+            )
+
+            if combined_audio is None:
+                combined_audio = clip_segment
+            else:
+                combined_audio += clip_segment
+
+            if i < len(clips_to_join) - 1:
+                pause_ms = clip_info.get("pause_after_ms", default_pause_ms)
+                if pause_ms > 0:
+                    pause_segment = AudioSegment.silent(duration=pause_ms)
+                    combined_audio += pause_segment
+                    log.debug(
+                        f"{log_identifier} Added {pause_ms}ms pause after '{clip_filename_with_version}'."
+                    )
+
+        if combined_audio is None:
+            return {
+                "status": "error",
+                "message": "No audio clips were successfully processed.",
+            }
+
+        output_buffer = io.BytesIO()
+        await asyncio.to_thread(combined_audio.export, output_buffer, format="mp3")
+        mp3_data = output_buffer.getvalue()
+
+        final_filename = output_filename
+        if not final_filename:
+            final_filename = f"concatenated_audio_{uuid.uuid4()}.mp3"
+        elif not final_filename.lower().endswith(".mp3"):
+            final_filename = f"{final_filename}.mp3"
+
+        metadata = {
+            "description": f"Concatenated audio created from {len(clips_to_join)} clips.",
+            "source_clips": source_filenames,
+            "generation_tool": "concatenate_audio",
+            "generation_timestamp": datetime.now(timezone.utc).isoformat(),
+        }
+
+        log.info(
+            f"{log_identifier} Saving concatenated audio artifact '{final_filename}'"
+        )
+        save_result = await _save_audio_artifact(
+            audio_data=mp3_data,
+            filename=final_filename,
+            metadata=metadata,
+            tool_context=tool_context,
+        )
+
+        log.info(
+            f"{log_identifier} Concatenated audio artifact saved successfully: {final_filename} v{save_result['data_version']}"
+        )
+
+        return {
+            "status": "success",
+            "message": f"Audio clips concatenated and saved successfully.",
+            "output_filename": final_filename,
+            "output_version": save_result["data_version"],
+            "result_preview": f"Concatenated audio '{final_filename}' (v{save_result['data_version']}) created from {len(clips_to_join)} clips.",
+        }
+
+    except FileNotFoundError as e:
+        log.warning(f"{log_identifier} File not found error: {e}")
+        return {"status": "error", "message": str(e)}
+    except ValueError as ve:
+        log.error(f"{log_identifier} Value error: {ve}")
+        return {"status": "error", "message": str(ve)}
+    except IOError as ioe:
+        log.error(f"{log_identifier} IO error: {ioe}")
+        return {"status": "error", "message": str(ioe)}
+    except Exception as e:
+        log.exception(f"{log_identifier} Unexpected error in concatenate_audio: {e}")
+        return {"status": "error", "message": f"An unexpected error occurred: {e}"}
+
+
+async def transcribe_audio(
+    audio_filename: str,
+    tool_context: ToolContext = None,
+    tool_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """
+    Transcribes an audio recording using an OpenAI-compatible audio transcription API.
+
+    Args:
+        audio_filename: The filename (and optional :version) of the input audio artifact.
+        tool_context: The context provided by the ADK framework.
+        tool_config: Configuration dictionary containing model, api_base, api_key.
+
+    Returns:
+        A dictionary containing:
+        - "status": "success" or "error".
+        - "message": A descriptive message about the outcome.
+        - "transcription": The transcribed text from the API (if successful).
+        - "audio_filename": The name of the input audio artifact (if successful).
+        - "audio_version": The version of the input audio artifact (if successful).
+    """
+    log_identifier = f"[AudioTools:transcribe_audio:{audio_filename}]"
+    if not tool_context:
+        log.error(f"{log_identifier} ToolContext is missing.")
+        return {"status": "error", "message": "ToolContext is missing."}
+
+    try:
+        inv_context = tool_context._invocation_context
+        if not inv_context:
+            raise ValueError("InvocationContext is not available.")
+
+        app_name = getattr(inv_context, "app_name", None)
+        user_id = getattr(inv_context, "user_id", None)
+        session_id = get_original_session_id(inv_context)
+        artifact_service = getattr(inv_context, "artifact_service", None)
+
+        if not all([app_name, user_id, session_id, artifact_service]):
+            missing_parts = [
+                part
+                for part, val in [
+                    ("app_name", app_name),
+                    ("user_id", user_id),
+                    ("session_id", session_id),
+                    ("artifact_service", artifact_service),
+                ]
+                if not val
+            ]
+            raise ValueError(
+                f"Missing required context parts: {', '.join(missing_parts)}"
+            )
+
+        log.info(f"{log_identifier} Processing request for session {session_id}.")
+
+        current_tool_config = tool_config if tool_config is not None else {}
+
+        if not current_tool_config:
+            log.warning(
+                f"{log_identifier} Tool-specific configuration (tool_config) is empty."
+            )
+
+        model_name = current_tool_config.get("model")
+        api_key = current_tool_config.get("api_key")
+        api_base = current_tool_config.get("api_base")
+
+        if not model_name:
+            raise ValueError("'model' configuration is missing in tool_config.")
+        if not api_key:
+            raise ValueError("'api_key' configuration is missing in tool_config.")
+        if not api_base:
+            raise ValueError("'api_base' configuration is missing in tool_config.")
+
+        log.debug(f"{log_identifier} Using model: {model_name}, API base: {api_base}")
+
+        parts = audio_filename.split(":", 1)
+        filename_base_for_load = parts[0]
+        version_str = parts[1] if len(parts) > 1 else None
+        version_to_load = int(version_str) if version_str else None
+
+        if not _is_supported_audio_format_for_transcription(filename_base_for_load):
+            raise ValueError(f"Unsupported audio format. Supported formats: .wav, .mp3")
+
+        if version_to_load is None:
+            list_versions_method = getattr(artifact_service, "list_versions")
+            if inspect.iscoroutinefunction(list_versions_method):
+                versions = await list_versions_method(
+                    app_name=app_name,
+                    user_id=user_id,
+                    session_id=session_id,
+                    filename=filename_base_for_load,
+                )
+            else:
+                versions = await asyncio.to_thread(
+                    list_versions_method,
+                    app_name=app_name,
+                    user_id=user_id,
+                    session_id=session_id,
+                    filename=filename_base_for_load,
+                )
+            if not versions:
+                raise FileNotFoundError(
+                    f"Audio artifact '{filename_base_for_load}' not found."
+                )
+            version_to_load = max(versions)
+            log.debug(
+                f"{log_identifier} Using latest version for input: {version_to_load}"
+            )
+
+        load_artifact_method = getattr(artifact_service, "load_artifact")
+        if inspect.iscoroutinefunction(load_artifact_method):
+            audio_artifact_part = await load_artifact_method(
+                app_name=app_name,
+                user_id=user_id,
+                session_id=session_id,
+                filename=filename_base_for_load,
+                version=version_to_load,
+            )
+        else:
+            audio_artifact_part = await asyncio.to_thread(
+                load_artifact_method,
+                app_name=app_name,
+                user_id=user_id,
+                session_id=session_id,
+                filename=filename_base_for_load,
+                version=version_to_load,
+            )
+
+        if not audio_artifact_part or not audio_artifact_part.inline_data:
+            raise FileNotFoundError(
+                f"Content for audio artifact '{filename_base_for_load}' v{version_to_load} not found."
+            )
+
+        audio_bytes = audio_artifact_part.inline_data.data
+        log.debug(f"{log_identifier} Loaded audio artifact: {len(audio_bytes)} bytes")
+
+        temp_file_path = None
+        try:
+            file_ext = os.path.splitext(filename_base_for_load)[1]
+
+            with tempfile.NamedTemporaryFile(
+                suffix=file_ext, delete=False
+            ) as temp_file:
+                temp_file_path = temp_file.name
+                temp_file.write(audio_bytes)
+
+            log.debug(f"{log_identifier} Created temporary file: {temp_file_path}")
+
+            api_url = f"{api_base.rstrip('/')}/v1/audio/transcriptions"
+            headers = {"Authorization": f"Bearer {api_key}"}
+
+            mime_type = _get_audio_mime_type(filename_base_for_load)
+
+            with open(temp_file_path, "rb") as audio_file:
+                files = {
+                    "file": (filename_base_for_load, audio_file, mime_type),
+                    "model": (None, model_name),
+                }
+
+                log.debug(f"{log_identifier} Calling transcription API...")
+
+                async with httpx.AsyncClient(timeout=60.0) as client:
+                    response = await client.post(api_url, headers=headers, files=files)
+                    response.raise_for_status()
+                    response_data = response.json()
+
+            log.debug(f"{log_identifier} Transcription API response received.")
+
+            if not response_data.get("text"):
+                raise ValueError("API response does not contain transcription text.")
+
+            transcription = response_data["text"]
+
+            log.info(
+                f"{log_identifier} Audio transcribed successfully. Transcription length: {len(transcription)} characters"
+            )
+
+            return {
+                "status": "success",
+                "message": "Audio transcribed successfully",
+                "transcription": transcription,
+                "audio_filename": filename_base_for_load,
+                "audio_version": version_to_load,
+            }
+
+        finally:
+            if temp_file_path and os.path.exists(temp_file_path):
+                try:
+                    os.remove(temp_file_path)
+                    log.debug(
+                        f"{log_identifier} Cleaned up temporary file: {temp_file_path}"
+                    )
+                except OSError as e:
+                    log.warning(
+                        f"{log_identifier} Failed to clean up temporary file {temp_file_path}: {e}"
+                    )
+
+    except FileNotFoundError as e:
+        log.warning(f"{log_identifier} File not found error: {e}")
+        return {"status": "error", "message": str(e)}
+    except ValueError as ve:
+        log.error(f"{log_identifier} Value error: {ve}")
+        return {"status": "error", "message": str(ve)}
+    except httpx.HTTPStatusError as hse:
+        log.error(
+            f"{log_identifier} HTTP error calling transcription API: {hse.response.status_code} - {hse.response.text}"
+        )
+        return {"status": "error", "message": f"API error: {hse.response.status_code}"}
+    except httpx.RequestError as re:
+        log.error(f"{log_identifier} Request error calling transcription API: {re}")
+        return {"status": "error", "message": f"Request error: {re}"}
+    except json.JSONDecodeError as jde:
+        log.error(f"{log_identifier} JSON decode error: {jde}")
+        return {"status": "error", "message": "Invalid JSON response from API"}
+    except Exception as e:
+        log.exception(f"{log_identifier} Unexpected error in transcribe_audio: {e}")
+        return {"status": "error", "message": f"An unexpected error occurred: {e}"}
+
+
+select_voice_tool_def = BuiltinTool(
+    name="select_voice",
+    implementation=select_voice,
+    description="Selects a suitable voice name based on criteria like gender and tone. Use this to get a consistent voice name that can be passed to the `text_to_speech` tool for multiple calls.",
+    category="audio",
+    required_scopes=["tool:audio:tts"],
+    parameters=adk_types.Schema(
+        type=adk_types.Type.OBJECT,
+        properties={
+            "gender": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional desired gender for the voice ('male', 'female', 'neutral').",
+                nullable=True,
+            ),
+            "tone": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional tone preference (e.g., 'friendly', 'professional').",
+                nullable=True,
+            ),
+            "exclude_voices": adk_types.Schema(
+                type=adk_types.Type.ARRAY,
+                items=adk_types.Schema(type=adk_types.Type.STRING),
+                description="Optional list of voice names to exclude from the selection.",
+                nullable=True,
+            ),
+        },
+        required=[],
+    ),
+    examples=[],
+)
+
+text_to_speech_tool_def = BuiltinTool(
+    name="text_to_speech",
+    implementation=text_to_speech,
+    description="Converts text to speech using Gemini TTS API and saves as MP3 artifact.",
+    category="audio",
+    required_scopes=["tool:audio:tts"],
+    parameters=adk_types.Schema(
+        type=adk_types.Type.OBJECT,
+        properties={
+            "text": adk_types.Schema(
+                type=adk_types.Type.STRING, description="The text to convert to speech."
+            ),
+            "output_filename": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional filename for the output MP3.",
+                nullable=True,
+            ),
+            "voice_name": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional specific voice name (e.g., 'Kore', 'Puck'). Overrides gender and tone.",
+                nullable=True,
+            ),
+            "gender": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional desired gender for the voice ('male', 'female', 'neutral'). Used if 'voice_name' is not provided.",
+                nullable=True,
+            ),
+            "tone": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional tone preference (e.g., 'friendly', 'professional'). Used if 'voice_name' is not provided.",
+                nullable=True,
+            ),
+            "language": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional language code (e.g., 'en-US', 'fr-FR').",
+                nullable=True,
+            ),
+        },
+        required=["text"],
+    ),
+    examples=[],
+)
+
+multi_speaker_text_to_speech_tool_def = BuiltinTool(
+    name="multi_speaker_text_to_speech",
+    implementation=multi_speaker_text_to_speech,
+    description="Converts conversation text with speaker labels to speech using multiple voices.",
+    category="audio",
+    required_scopes=["tool:audio:tts"],
+    parameters=adk_types.Schema(
+        type=adk_types.Type.OBJECT,
+        properties={
+            "conversation_text": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Text with speaker labels (e.g., 'Speaker1: Hello\\nSpeaker2: Hi there').",
+            ),
+            "output_filename": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional filename for the output MP3.",
+                nullable=True,
+            ),
+            "speaker_configs": adk_types.Schema(
+                type=adk_types.Type.ARRAY,
+                items=adk_types.Schema(
+                    type=adk_types.Type.OBJECT,
+                    description="Configuration for a single speaker.",
+                    properties={
+                        "name": adk_types.Schema(
+                            type=adk_types.Type.STRING,
+                            description="The speaker's name as it appears in conversation_text.",
+                        ),
+                        "voice": adk_types.Schema(
+                            type=adk_types.Type.STRING,
+                            description="Specific voice name to use. Overrides gender/tone.",
+                            nullable=True,
+                        ),
+                        "gender": adk_types.Schema(
+                            type=adk_types.Type.STRING,
+                            description="Desired gender ('male', 'female', 'neutral').",
+                            nullable=True,
+                        ),
+                        "tone": adk_types.Schema(
+                            type=adk_types.Type.STRING,
+                            description="Desired tone (e.g., 'friendly').",
+                            nullable=True,
+                        ),
+                    },
+                ),
+                description="Optional list of speaker configurations.",
+                nullable=True,
+            ),
+            "language": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional language code (e.g., 'en-US', 'fr-FR').",
+                nullable=True,
+            ),
+        },
+        required=["conversation_text"],
+    ),
+    examples=[],
+)
+
+concatenate_audio_tool_def = BuiltinTool(
+    name="concatenate_audio",
+    implementation=concatenate_audio,
+    description="Combines multiple audio artifacts in a specified order into a single audio file. Allows for custom pause durations between each clip.",
+    category="audio",
+    required_scopes=["tool:audio:edit", "tool:artifact:create", "tool:artifact:load"],
+    parameters=adk_types.Schema(
+        type=adk_types.Type.OBJECT,
+        properties={
+            "clips_to_join": adk_types.Schema(
+                type=adk_types.Type.ARRAY,
+                description="An ordered list of clip objects to be joined.",
+                items=adk_types.Schema(
+                    type=adk_types.Type.OBJECT,
+                    properties={
+                        "filename": adk_types.Schema(
+                            type=adk_types.Type.STRING,
+                            description="The artifact filename of the audio clip (with optional :version).",
+                        ),
+                        "pause_after_ms": adk_types.Schema(
+                            type=adk_types.Type.INTEGER,
+                            description="Optional duration of silence in milliseconds to insert *after* this clip. Defaults to 500ms.",
+                            nullable=True,
+                        ),
+                    },
+                    required=["filename"],
+                ),
+            ),
+            "output_filename": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="Optional. The desired filename for the final combined audio artifact.",
+                nullable=True,
+            ),
+        },
+        required=["clips_to_join"],
+    ),
+    examples=[],
+)
+
+transcribe_audio_tool_def = BuiltinTool(
+    name="transcribe_audio",
+    implementation=transcribe_audio,
+    description="Transcribes an audio recording using an OpenAI-compatible audio transcription API.",
+    category="audio",
+    required_scopes=["tool:audio:transcribe"],
+    parameters=adk_types.Schema(
+        type=adk_types.Type.OBJECT,
+        properties={
+            "audio_filename": adk_types.Schema(
+                type=adk_types.Type.STRING,
+                description="The filename (and optional :version) of the input audio artifact.",
+            ),
+        },
+        required=["audio_filename"],
+    ),
+    examples=[],
+)
+
+tool_registry.register(select_voice_tool_def)
+tool_registry.register(text_to_speech_tool_def)
+tool_registry.register(multi_speaker_text_to_speech_tool_def)
+tool_registry.register(concatenate_audio_tool_def)
+tool_registry.register(transcribe_audio_tool_def)