All TTS vendors tracked by CodeSOTA.

Cartesia, ElevenLabs Flash, Gemini Flash, Gradium

Prioritize TTFB, streaming stability and interruption handling.

ElevenLabs, PlayHT, Google Chirp 3, Gradium, Coqui

Check consent, similarity, speaker consistency and policy terms.

Sesame, Kokoro, Orpheus, F5-TTS, Dia, Fish Speech

Validate license, hardware footprint, language coverage and serving stack.

OpenAI, Google, ElevenLabs

Use when integration speed and vendor support beat fine-grained control.

ElevenLabs

Commercial TTS quality leader with a large voice library and cloning workflow.

Best buyer

Creators, media products and apps that need polished voices quickly.

Strengths

top MOSvoice librarycloning

Verify

Cost, data policy, latency tier and voice-rights workflow matter at scale.

Tracked models

ElevenLabs Turbo v2.5 (4.8 MOS), ElevenLabs Flash v2.5 (4.6 MOS)

Cartesia

Low-latency hosted TTS for voice agents and conversational interfaces.

Best buyer

Product teams building real-time phone, agent, avatar or interruptible voice systems.

Strengths

fast first bytestreamingagent fit

Verify

Benchmark naturalness is only part of the decision; test interruption handling and regional latency.

Tracked models

Cartesia Sonic 2 (4.7 MOS)

Google

Hyperscaler TTS with Gemini-native audio and Cloud Text-to-Speech coverage.

Best buyer

Enterprises already on Google Cloud or products needing many locales and managed operations.

Strengths

locale coveragecloud operationsprompted style

Verify

Quality varies by voice, language and product family; compare Chirp, Gemini and classic Cloud TTS separately.

Tracked models

Gemini 2.5 Pro TTS (4.7 MOS), Gemini 2.5 Flash TTS (4.5 MOS), Google Chirp 3 HD (4.4 MOS)

OpenAI

Simple hosted TTS API for teams already using OpenAI infrastructure.

Best buyer

Developers who want quick integration, stable docs and a small set of built-in voices.

Strengths

simple APImanaged servicedeveloper adoption

Verify

Voice range, style control and cloning depth are narrower than specialist TTS platforms.

Tracked models

OpenAI TTS HD (4.7 MOS)

PlayHT

Hosted voice generation platform focused on cloning and creator workflows.

Best buyer

Media, marketing and product teams that need cloned or branded voices without self-hosting.

Strengths

voice cloningemotion controlhosted workflow

Verify

Compare consent workflow, long-form consistency and cost before large-volume use.

Tracked models

PlayHT 3.0 (4.6 MOS)

Gradium

Hosted TTS API tracked by CodeSOTA for intelligibility, UTMOS and first-byte latency.

Best buyer

Voice-agent teams that care about streaming, cloning and measured information preservation.

Strengths

streamingvoice cloningCodeSOTA measured

Verify

Run your own hard-prompt set for names, numbers, URLs and domain vocabulary.

Tracked models

Gradium TTS (4.4 MOS)

Sesame

Open conversational speech model with expressive dialogue quality.

Best buyer

Teams evaluating open alternatives to commercial conversational TTS.

Strengths

conversationexpressivenessopen source

Verify

Measure deployment latency and license fit before treating it as a hosted-vendor replacement.

Tracked models

Sesame CSM (4.7 MOS)

Canopy Labs

Open LLM-style speech generation through Orpheus TTS.

Best buyer

Teams that want an expressive open model they can fine-tune or inspect.

Strengths

emotion tagscustom fine-tuningopen deployment

Verify

Operational burden is on you: serving, latency tuning, voice QA and license review.

Tracked models

Orpheus TTS (4.6 MOS)

Fish Audio

Open multilingual speech generation with strong CJK coverage.

Best buyer

Teams experimenting with self-hosted multilingual or regional-language TTS.

Strengths

multilingualCJK supportopen weights

Verify

Validate English naturalness, serving cost and license terms against your exact product use.

Tracked models

Fish Audio S2 Pro (4.6 MOS), Fish Speech 1.5 (4.4 MOS)

Coqui

Open-source TTS toolkit with XTTS voice cloning as the practical draw.

Best buyer

Developers who need self-hosted multilingual cloning and can manage model infrastructure.

Strengths

voice cloning17 languageslocal control

Verify

Project and license status need checking before commercial deployment.

Tracked models

XTTS v2 (4.5 MOS)

Hexgrad

Tiny open TTS through Kokoro: strong quality for an 82M parameter model.

Best buyer

Builders who need lightweight local TTS, edge experiments or cheap batch generation.

Strengths

small modelCPU friendlyApache 2.0

Verify

Not a full hosted vendor; you still need serving, voice management and QA.

Tracked models

Kokoro v1.0 (4.5 MOS)

Shanghai AI Lab

Research-grade open zero-shot cloning through F5-TTS.

Best buyer

Research and engineering teams testing fast open cloning and adaptation.

Strengths

flow matchingzero-shot cloningresearch code

Verify

Needs production hardening: model serving, monitoring, safety and voice-rights controls.

Tracked models

F5-TTS (4.4 MOS)

Nari Labs

Dialogue-focused open TTS with non-verbal cues through Dia.

Best buyer

Teams prototyping expressive dialogue, characters or conversational audio.

Strengths

dialoguenon-verbal cuesopen model

Verify

Treat as a model profile, not a managed API; validate consistency over long sessions.

Tracked models

Dia 1.6B (4.3 MOS)

SparkAudio

Open controllable TTS with explicit pitch, speed and emotion attributes.

Best buyer

Teams that need controllable generation experiments rather than a black-box API.

Strengths

attribute controlmultilingualopen model

Verify

Validate actual controllability and stability across speakers before product use.

Tracked models

Spark-TTS (4.3 MOS)

Supertone

Compact local multilingual TTS through Supertonic 3.

Best buyer

Teams testing on-device or edge speech generation with a small open-weight model.

Strengths

small footprintlocal inferencemultilingual

Verify

Add same-prompt listening samples and hard-text runs before ranking it against measured rows.

Tracked models

Supertonic 3 (4.2 MOS)

Hugging Face

Open prompt-controlled TTS through Parler-TTS and the wider HF ecosystem.

Best buyer

Research and prototype teams that want open training artifacts and model iteration.

Strengths

prompted styleopen toolingecosystem

Verify

Production reliability depends on your chosen hosting path and model variant.

Tracked models

Parler-TTS (4.1 MOS)

Rhasspy

Lightweight local TTS through Piper for embedded and offline systems.

Best buyer

Home assistant, robotics, kiosk and edge teams that need fast local speech.

Strengths

offlineRaspberry Pi fitlow latency

Verify

Naturalness trails frontier hosted voices; choose it for control and footprint.

Tracked models

Piper (3.6 MOS)