Text-to-Speech

Text → audio bytes. Same three providers, three SDK calls, three big voice catalogs. Every voice ID below was verified against the live docs.

Official docs verified 2026-08-08

• Azure Speech TTS: learn.microsoft.com/…/speech-service/get-started-text-to-speech
• GCP TTS SDK: cloud.google.com/text-to-speech/docs/quickstart-client-libraries
• GCP TTS voice list: cloud.google.com/text-to-speech/docs/list-voices-and-types
• AWS Polly SDK: docs.aws.amazon.com/polly/…/synthesize-speech-python
• AWS Polly voice catalog: docs.aws.amazon.com/polly/…/available-voices

Voice IDs (English US, verified)

Provider	Voice ID	Kind	Notes
Azure Speech	en-US-Ava:DragonHDLatestNeural	HD Neural	Newest voice family.
Azure Speech	en-US-JennyNeural	Neural	Long-standing default.
Azure Speech	en-US-AndrewMultilingualNeural	Multilingual Neural	Cross-language reuse.
GCP TTS	en-US-Chirp3-HD-Achernar (F), en-US-Chirp3-HD-Fenrir (M)	Chirp 3 HD	Newest voice tier.
GCP TTS	en-US-Neural2-C (F), en-US-Neural2-A (M)	Neural2	Premium standard.
AWS Polly	Ruth / Amy / Matthew / Danielle	Generative	Engine="generative".
AWS Polly	Joanna / Salli / Kevin / Kimberly	Neural	Engine="neural".

The full catalogs are large and drift often — hit the docs URLs above before pinning to production.

Audio formats

Provider	Default output	Programmatic override
Azure Speech	24 kHz 48 kb/s mono MP3	SpeechConfig.set_speech_synthesis_output_format(...)
GCP TTS	Chosen per call	AudioConfig(audio_encoding=AudioEncoding.MP3) / .LINEAR16 / .OGG_OPUS
AWS Polly	Chosen per call	OutputFormat="mp3" / "pcm" / "ogg_vorbis" / "json"

The examples default to MP3 so the raw bytes are directly playable.

Native per-cloud calls

AzureGCPAWS

import os
import azure.cognitiveservices.speech as speechsdk

speech_config = speechsdk.SpeechConfig(
    subscription=os.environ["SPEECH_KEY"],
    endpoint=os.environ["SPEECH_ENDPOINT"],
)
speech_config.speech_synthesis_voice_name = os.environ.get(
    "AZURE_TTS_VOICE", "en-US-JennyNeural",
)
# Get bytes back instead of playing to the local speaker.
audio_out = speechsdk.audio.PullAudioOutputStream()
audio_config = speechsdk.audio.AudioOutputConfig(stream=audio_out)
synth = speechsdk.SpeechSynthesizer(
    speech_config=speech_config, audio_config=audio_config,
)
result = synth.speak_text_async("Hello from Azure Speech").get()
audio_bytes = result.audio_data

import os
from google.cloud import texttospeech

client = texttospeech.TextToSpeechClient()
resp = client.synthesize_speech(
    input=texttospeech.SynthesisInput(text="Hello from Cloud TTS"),
    voice=texttospeech.VoiceSelectionParams(
        language_code="en-US",
        name=os.environ.get("GCP_TTS_VOICE", "en-US-Chirp3-HD-Achernar"),
    ),
    audio_config=texttospeech.AudioConfig(
        audio_encoding=texttospeech.AudioEncoding.MP3,
    ),
)
audio_bytes = resp.audio_content

import os, boto3

client = boto3.client("polly", region_name=os.environ.get("AWS_REGION", "us-east-1"))
resp = client.synthesize_speech(
    Text="Hello from Polly",
    VoiceId=os.environ.get("AWS_POLLY_VOICE", "Ruth"),
    OutputFormat="mp3",
    Engine=os.environ.get("AWS_POLLY_ENGINE", "generative"),
)
audio_bytes = resp["AudioStream"].read()

A common signature

def synthesize(text: str) -> bytes: ...

Returns MP3 bytes by default; the voice service writes them straight to the HTTP response.