tts

Sarvam AI text-to-speech service implementation.

This module provides TTS services using Sarvam AI’s API with support for multiple Indian languages and two model variants:

Model Variants:

• bulbul:v2 (default): Standard TTS model

  • Supports: pitch, loudness, pace (0.3-3.0)

  • Default sample rate: 22050 Hz

  • Speakers: anushka (default), abhilash, manisha, vidya, arya, karun, hitesh

• bulbul:v3-beta: Advanced TTS model with temperature control

  • Does NOT support: pitch, loudness

  • Supports: pace (0.5-2.0), temperature (0.01-1.0)

  • Default sample rate: 24000 Hz

  • Preprocessing is always enabled

  • Speakers: aditya (default), ritu, priya, neha, rahul, pooja, rohan, simran, kavya, amit, dev, ishita, shreya, ratan, varun, manan, sumit, roopa, kabir, aayan, shubh, ashutosh, advait, amelia, sophia

• bulbul:v3: Advanced TTS model with temperature control

  • Does NOT support: pitch, loudness

  • Supports: pace (0.5-2.0), temperature (0.01-1.0)

  • Default sample rate: 24000 Hz

  • Preprocessing is always enabled

  • Speakers: aditya (default), ritu, priya, neha, rahul, pooja, rohan, simran, kavya, amit, dev, ishita, shreya, ratan, varun, manan, sumit, roopa, kabir, aayan, shubh, ashutosh, advait, amelia, sophia

See https://docs.sarvam.ai/api-reference-docs/text-to-speech/stream for full API details.

class pipecat.services.sarvam.tts.SarvamTTSModel(*values)

Bases: StrEnum

Available Sarvam TTS models.

Parameters:

• BULBUL_V2 – Standard TTS model with pitch/loudness control. - Supports pitch, loudness, pace (0.3-3.0) - Default sample rate: 22050 Hz

• BULBUL_V3_BETA – Advanced model with temperature control. - Does NOT support pitch/loudness - Pace range: 0.5-2.0 - Supports temperature parameter - Default sample rate: 24000 Hz - Preprocessing is always enabled

BULBUL_V2 = 'bulbul:v2'

BULBUL_V3_BETA = 'bulbul:v3-beta'

BULBUL_V3 = 'bulbul:v3'

class pipecat.services.sarvam.tts.SarvamTTSSpeakerV2(*values)

Bases: StrEnum

Available speakers for bulbul:v2 model.

Female voices: anushka, manisha, vidya, arya Male voices: abhilash, karun, hitesh

ANUSHKA = 'anushka'

ABHILASH = 'abhilash'

MANISHA = 'manisha'

VIDYA = 'vidya'

ARYA = 'arya'

KARUN = 'karun'

HITESH = 'hitesh'

class pipecat.services.sarvam.tts.SarvamTTSSpeakerV3(*values)

Bases: StrEnum

Available speakers for bulbul:v3-beta model.

Includes a wider variety of voices with different characteristics.

ADITYA = 'aditya'

RITU = 'ritu'

PRIYA = 'priya'

NEHA = 'neha'

RAHUL = 'rahul'

POOJA = 'pooja'

ROHAN = 'rohan'

SIMRAN = 'simran'

KAVYA = 'kavya'

AMIT = 'amit'

DEV = 'dev'

ISHITA = 'ishita'

SHREYA = 'shreya'

RATAN = 'ratan'

VARUN = 'varun'

MANAN = 'manan'

SUMIT = 'sumit'

ROOPA = 'roopa'

KABIR = 'kabir'

AAYAN = 'aayan'

SHUBH = 'shubh'

ASHUTOSH = 'ashutosh'

ADVAIT = 'advait'

AMELIA = 'amelia'

SOPHIA = 'sophia'

class pipecat.services.sarvam.tts.TTSModelConfig(supports_pitch: bool, supports_loudness: bool, supports_temperature: bool, default_sample_rate: int, default_speaker: str, pace_range: tuple[float, float], preprocessing_always_enabled: bool, speakers: tuple[str, ...])

Bases: object

Immutable configuration for a Sarvam TTS model.

Parameters:

• supports_pitch – Whether the model accepts pitch parameter.

• supports_loudness – Whether the model accepts loudness parameter.

• supports_temperature – Whether the model accepts temperature parameter.

• default_sample_rate – Default audio sample rate in Hz.

• default_speaker – Default speaker voice ID.

• pace_range – Valid range for pace parameter (min, max).

• preprocessing_always_enabled – Whether preprocessing is always enabled.

• speakers – Tuple of available speaker names for this model.

supports_pitch: bool

supports_loudness: bool

supports_temperature: bool

default_sample_rate: int

default_speaker: str

pace_range: tuple[float, float]

preprocessing_always_enabled: bool

speakers: tuple[str, ...]

pipecat.services.sarvam.tts.get_speakers_for_model(model: str) → list[str]

Get the list of available speakers for a given model.

Parameters:

model – The model name (e.g., “bulbul:v2” or “bulbul:v3-beta”).

Returns:

List of speaker names available for the model.

pipecat.services.sarvam.tts.language_to_sarvam_language(language: Language) → str | None

Convert Pipecat Language enum to Sarvam AI language codes.

Parameters:

language – The Language enum value to convert.

Returns:

The corresponding Sarvam AI language code, or None if not supported.

class pipecat.services.sarvam.tts.SarvamHttpTTSSettings(model: str | None | _NotGiven = <factory>, extra: dict[str, ~typing.Any]=<factory>, voice: str | None | _NotGiven = <factory>, language: Language | str | None | _NotGiven = <factory>, enable_preprocessing: bool | None | _NotGiven = <factory>, pace: float | None | _NotGiven = <factory>, pitch: float | None | _NotGiven = <factory>, loudness: float | None | _NotGiven = <factory>, temperature: float | None | _NotGiven = <factory>)

Bases: TTSSettings

Settings for SarvamHttpTTSService.

Parameters:

• enable_preprocessing – Whether to enable text preprocessing. Defaults to False. Note: Always enabled for bulbul:v3-beta (cannot be disabled).

• pace – Speech pace multiplier. Defaults to 1.0. - bulbul:v2: Range 0.3 to 3.0 - bulbul:v3-beta: Range 0.5 to 2.0

• pitch – Voice pitch adjustment (-0.75 to 0.75). Defaults to 0.0. Note: Only supported for bulbul:v2. Ignored for v3 models.

• loudness – Volume multiplier (0.3 to 3.0). Defaults to 1.0. Note: Only supported for bulbul:v2. Ignored for v3 models.

• temperature – Controls output randomness for bulbul:v3-beta (0.01 to 1.0). Lower values = more deterministic, higher = more random. Defaults to 0.6. Note: Only supported for bulbul:v3-beta. Ignored for v2.

enable_preprocessing: bool | None | _NotGiven

pace: float | None | _NotGiven

pitch: float | None | _NotGiven

loudness: float | None | _NotGiven

temperature: float | None | _NotGiven

class pipecat.services.sarvam.tts.SarvamTTSSettings(model: str | None | _NotGiven = <factory>, extra: dict[str, ~typing.Any]=<factory>, voice: str | None | _NotGiven = <factory>, language: Language | str | None | _NotGiven = <factory>, enable_preprocessing: bool | None | _NotGiven = <factory>, pace: float | None | _NotGiven = <factory>, pitch: float | None | _NotGiven = <factory>, loudness: float | None | _NotGiven = <factory>, temperature: float | None | _NotGiven = <factory>, min_buffer_size: int | None | _NotGiven = <factory>, max_chunk_length: int | None | _NotGiven = <factory>)

Bases: SarvamHttpTTSSettings

Settings for SarvamTTSService.

Extends SarvamHttpTTSService.Settings with WebSocket-specific buffering parameters.

Parameters:

• min_buffer_size – Minimum characters to buffer before generating audio. Lower values reduce latency but may affect quality. Defaults to 50.

• max_chunk_length – Maximum characters processed in a single chunk. Controls memory usage and processing efficiency. Defaults to 150.

min_buffer_size: int | None | _NotGiven

max_chunk_length: int | None | _NotGiven

class pipecat.services.sarvam.tts.SarvamHttpTTSService(*, api_key: str, aiohttp_session: ClientSession, voice_id: str | None = None, model: str | None = None, base_url: str = 'https://api.sarvam.ai', sample_rate: int | None = None, params: InputParams | None = None, settings: SarvamHttpTTSSettings | None = None, **kwargs)

Bases: TTSService

Text-to-Speech service using Sarvam AI’s API.

Converts text to speech using Sarvam AI’s TTS models with support for multiple Indian languages. Provides control over voice characteristics.

Model Differences:

• bulbul:v2 (default):

  • Supports: pitch (-0.75 to 0.75), loudness (0.3 to 3.0), pace (0.3 to 3.0)

  • Default sample rate: 22050 Hz

  • Speakers: anushka, abhilash, manisha, vidya, arya, karun, hitesh

• bulbul:v3-beta:

  • Does NOT support: pitch, loudness (will be ignored)

  • Supports: pace (0.5 to 2.0), temperature (0.01 to 1.0)

  • Default sample rate: 24000 Hz

  • Preprocessing is always enabled

  • Speakers: aditya, ritu, priya, neha, rahul, pooja, rohan, simran, kavya, amit, dev, ishita, shreya, ratan, varun, manan, sumit, roopa, kabir, aayan, shubh, ashutosh, advait, amelia, sophia

Example:

# Using bulbul:v2 (default)
tts = SarvamHttpTTSService(
    api_key="your-api-key",
    aiohttp_session=session,
    settings=SarvamHttpTTSService.Settings(
        voice="anushka",
        model="bulbul:v2",
        language=Language.HI,
        pitch=0.1,
        pace=1.2,
        loudness=1.5,
    ),
)

# Using bulbul:v3-beta with temperature control
tts_v3 = SarvamHttpTTSService(
    api_key="your-api-key",
    aiohttp_session=session,
    settings=SarvamHttpTTSService.Settings(
        voice="aditya",  # Use v3 speaker
        model="bulbul:v3-beta",
        language=Language.HI,
        pace=1.2,  # Range: 0.5-2.0 for v3
        temperature=0.8,
    ),
)

Settings

alias of SarvamHttpTTSSettings

class InputParams(*, language: Language | None = Language.EN, pitch: Annotated[float | None, Ge(ge=-0.75), Le(le=0.75)] = 0.0, pace: Annotated[float | None, Ge(ge=0.3), Le(le=3.0)] = 1.0, loudness: Annotated[float | None, Ge(ge=0.3), Le(le=3.0)] = 1.0, enable_preprocessing: bool | None = False, temperature: Annotated[float | None, Ge(ge=0.01), Le(le=1.0)] = 0.6)

Bases: BaseModel

Input parameters for Sarvam TTS configuration.

Deprecated since version 0.0.105: Use SarvamHttpTTSService.Settings directly via the settings parameter instead.

Parameters:

• language – Language for synthesis. Defaults to English (India).

• pitch – Voice pitch adjustment (-0.75 to 0.75). Defaults to 0.0. Note: Only supported for bulbul:v2. Ignored for v3 models.

• pace – Speech pace multiplier. Defaults to 1.0. - bulbul:v2: Range 0.3 to 3.0 - bulbul:v3-beta: Range 0.5 to 2.0

• loudness – Volume multiplier (0.3 to 3.0). Defaults to 1.0. Note: Only supported for bulbul:v2. Ignored for v3 models.

• enable_preprocessing – Whether to enable text preprocessing. Defaults to False. Note: Always enabled for bulbul:v3-beta (cannot be disabled).

• temperature – Controls output randomness for bulbul:v3-beta (0.01 to 1.0). Lower values = more deterministic, higher = more random. Defaults to 0.6. Note: Only supported for bulbul:v3-beta. Ignored for v2.

language: Language | None

pitch: float | None

pace: float | None

loudness: float | None

enable_preprocessing: bool | None

temperature: float | None

__init__(*, api_key: str, aiohttp_session: ClientSession, voice_id: str | None = None, model: str | None = None, base_url: str = 'https://api.sarvam.ai', sample_rate: int | None = None, params: InputParams | None = None, settings: SarvamHttpTTSSettings | None = None, **kwargs)

Initialize the Sarvam TTS service.

Parameters:

• api_key – Sarvam AI API subscription key.

• aiohttp_session – Shared aiohttp session for making requests.

• voice_id –

  Speaker voice ID. If None, uses model-appropriate default.

  Deprecated since version 0.0.105: Use settings=SarvamHttpTTSService.Settings(voice=...) instead.

• model –

  TTS model to use. Options: - “bulbul:v2” (default): Standard model with pitch/loudness support - “bulbul:v3-beta”: Advanced model with temperature control

  Deprecated since version 0.0.105: Use settings=SarvamHttpTTSService.Settings(model=...) instead.

• base_url – Sarvam AI API base URL. Defaults to “https://api.sarvam.ai”.

• sample_rate – Audio sample rate in Hz (8000, 16000, 22050, 24000). If None, uses model-specific default.

• params –

  Additional voice and preprocessing parameters. If None, uses defaults.

  Deprecated since version 0.0.105: Use settings=SarvamHttpTTSService.Settings(...) instead.

• settings – Runtime-updatable settings. When provided alongside deprecated parameters, settings values take precedence.

• **kwargs – Additional arguments passed to parent TTSService.

can_generate_metrics() → bool

Check if this service can generate processing metrics.

Returns:

True, as Sarvam service supports metrics generation.

language_to_service_language(language: Language) → str | None

Convert a Language enum to Sarvam AI language format.

Parameters:

language – The language to convert.

Returns:

The Sarvam AI-specific language code, or None if not supported.

async start(frame: StartFrame)

Start the Sarvam TTS service.

Parameters:

frame – The start frame containing initialization parameters.

async run_tts(text: str, context_id: str) → AsyncGenerator[Frame | None, None]

Generate speech from text using Sarvam AI’s API.

Parameters:

• text – The text to synthesize into speech.

• context_id – The context ID for tracking audio frames.

Yields:

Frame – Audio frames containing the synthesized speech.

class pipecat.services.sarvam.tts.SarvamTTSService(*, api_key: str, model: str | None = None, voice_id: str | None = None, url: str = 'wss://api.sarvam.ai/text-to-speech/ws', aggregate_sentences: bool | None = None, text_aggregation_mode: TextAggregationMode | None = None, sample_rate: int | None = None, params: InputParams | None = None, settings: SarvamTTSSettings | None = None, **kwargs)

Bases: InterruptibleTTSService

WebSocket-based text-to-speech service using Sarvam AI.

Provides streaming TTS with real-time audio generation for multiple Indian languages. Uses WebSocket for low-latency streaming audio synthesis.

Model Differences:

• bulbul:v2 (default):

  • Supports: pitch (-0.75 to 0.75), loudness (0.3 to 3.0), pace (0.3 to 3.0)

  • Default sample rate: 22050 Hz

  • Speakers: anushka, abhilash, manisha, vidya, arya, karun, hitesh

• bulbul:v3-beta / bulbul:v3:

  • Does NOT support: pitch, loudness (will be ignored)

  • Supports: pace (0.5 to 2.0), temperature (0.01 to 1.0)

  • Default sample rate: 24000 Hz

  • Preprocessing is always enabled

  • Speakers: aditya, ritu, priya, neha, rahul, pooja, rohan, simran, kavya, amit, dev, ishita, shreya, ratan, varun, manan, sumit, roopa, kabir, aayan, shubh, ashutosh, advait, amelia, sophia

WebSocket Protocol: The service uses a WebSocket connection for real-time streaming. Messages include: - config: Initial configuration with voice settings - text: Text chunks for synthesis - flush: Signal to process remaining buffered text - ping: Keepalive signal

Example:

# Using bulbul:v2 (default)
tts = SarvamTTSService(
    api_key="your-api-key",
    settings=SarvamTTSService.Settings(
        voice="anushka",
        model="bulbul:v2",
        language=Language.HI,
        pitch=0.1,
        pace=1.2,
        loudness=1.5,
    ),
)

# Using bulbul:v3-beta with temperature control
tts_v3 = SarvamTTSService(
    api_key="your-api-key",
    settings=SarvamTTSService.Settings(
        voice="aditya",  # Use v3 speaker
        model="bulbul:v3-beta",
        language=Language.HI,
        pace=1.2,  # Range: 0.5-2.0 for v3
        temperature=0.8,
    ),
)

See https://docs.sarvam.ai/api-reference-docs/text-to-speech/stream for API details.

Settings

alias of SarvamTTSSettings

class InputParams(*, pitch: Annotated[float | None, Ge(ge=-0.75), Le(le=0.75)] = 0.0, pace: Annotated[float | None, Ge(ge=0.3), Le(le=3.0)] = 1.0, loudness: Annotated[float | None, Ge(ge=0.3), Le(le=3.0)] = 1.0, enable_preprocessing: bool | None = False, min_buffer_size: int | None = 50, max_chunk_length: int | None = 150, output_audio_codec: str | None = 'linear16', output_audio_bitrate: str | None = '128k', language: Language | None = Language.EN, temperature: Annotated[float | None, Ge(ge=0.01), Le(le=1.0)] = 0.6)

Bases: BaseModel

Configuration parameters for Sarvam TTS WebSocket service.

Deprecated since version 0.0.105: Use SarvamTTSService.Settings directly via the settings parameter instead.

Parameters:

• pitch – Voice pitch adjustment (-0.75 to 0.75). Defaults to 0.0. Note: Only supported for bulbul:v2. Ignored for v3 models.

• pace – Speech pace multiplier. Defaults to 1.0. - bulbul:v2: Range 0.3 to 3.0 - bulbul:v3-beta: Range 0.5 to 2.0

• loudness – Volume multiplier (0.3 to 3.0). Defaults to 1.0. Note: Only supported for bulbul:v2. Ignored for v3 models.

• enable_preprocessing – Enable text preprocessing. Defaults to False. Note: Always enabled for bulbul:v3-beta.

• min_buffer_size – Minimum characters to buffer before generating audio. Lower values reduce latency but may affect quality. Defaults to 50.

• max_chunk_length – Maximum characters processed in a single chunk. Controls memory usage and processing efficiency. Defaults to 150.

• output_audio_codec – Audio codec format. Options: linear16, mulaw, alaw, opus, flac, aac, wav, mp3. Defaults to “linear16”.

• output_audio_bitrate – Audio bitrate (32k, 64k, 96k, 128k, 192k). Defaults to “128k”.

• language – Target language for synthesis. Supports Indian languages.

• temperature – Controls output randomness for bulbul:v3-beta (0.01 to 1.0). Lower = more deterministic, higher = more random. Defaults to 0.6. Note: Only supported for bulbul:v3-beta. Ignored for v2.

Speakers by Model:

bulbul:v2:

• Female: anushka (default), manisha, vidya, arya

• Male: abhilash, karun, hitesh

bulbul:v3-beta:

• aditya (default), ritu, priya, neha, rahul, pooja, rohan, simran, kavya, amit, dev, ishita, shreya, ratan, varun, manan, sumit, roopa, kabir, aayan, shubh, ashutosh, advait, amelia, sophia

pitch: float | None

pace: float | None

loudness: float | None

enable_preprocessing: bool | None

min_buffer_size: int | None

max_chunk_length: int | None

output_audio_codec: str | None

output_audio_bitrate: str | None

language: Language | None

temperature: float | None

__init__(*, api_key: str, model: str | None = None, voice_id: str | None = None, url: str = 'wss://api.sarvam.ai/text-to-speech/ws', aggregate_sentences: bool | None = None, text_aggregation_mode: TextAggregationMode | None = None, sample_rate: int | None = None, params: InputParams | None = None, settings: SarvamTTSSettings | None = None, **kwargs)

Initialize the Sarvam TTS service with voice and transport configuration.

Parameters:

• api_key – Sarvam API key for authenticating TTS requests.

• model –

  TTS model to use. Options: - “bulbul:v2” (default): Standard model with pitch/loudness support - “bulbul:v3-beta”: Advanced model with temperature control

  Deprecated since version 0.0.105: Use settings=SarvamTTSService.Settings(model=...) instead.

• voice_id –

  Speaker voice ID. If None, uses model-appropriate default.

  Deprecated since version 0.0.105: Use settings=SarvamTTSService.Settings(voice=...) instead.

• url – WebSocket URL for the TTS backend (default production URL).

• aggregate_sentences –

  Deprecated. Use text_aggregation_mode instead.

  Deprecated since version 0.0.104: Use text_aggregation_mode instead.

• text_aggregation_mode – How to aggregate text before synthesis.

• sample_rate – Output audio sample rate in Hz (8000, 16000, 22050, 24000). If None, uses model-specific default.

• params –

  Optional input parameters to override defaults.

  Deprecated since version 0.0.105: Use settings=SarvamTTSService.Settings(...) instead.

• settings – Runtime-updatable settings. When provided alongside deprecated parameters, settings values take precedence.

• **kwargs – Arguments forwarded to InterruptibleTTSService.

See https://docs.sarvam.ai/api-reference-docs/text-to-speech/stream

can_generate_metrics() → bool

Check if this service can generate processing metrics.

Returns:

True, as Sarvam service supports metrics generation.

language_to_service_language(language: Language) → str | None

Convert a Language enum to Sarvam AI language format.

Parameters:

language – The language to convert.

Returns:

The Sarvam AI-specific language code, or None if not supported.

async start(frame: StartFrame)

Start the Sarvam TTS service.

Parameters:

frame – The start frame containing initialization parameters.

async stop(frame: EndFrame)

Stop the Sarvam TTS service.

Parameters:

frame – The end frame.

async cancel(frame: CancelFrame)

Cancel the Sarvam TTS service.

Parameters:

frame – The cancel frame.

async flush_audio(context_id: str | None = None)

Flush any pending audio synthesis by sending flush command.

async run_tts(text: str, context_id: str) → AsyncGenerator[Frame | None, None]

Generate speech audio frames from input text using Sarvam TTS.

Sends text over WebSocket for synthesis and yields corresponding audio or status frames.

Parameters:

• text – The text input to synthesize.

• context_id – The context ID for tracking audio frames.

Yields:

Frame objects including TTSStartedFrame, TTSAudioRawFrame(s, context_id=context_id), or TTSStoppedFrame.