text-to-speech synthesis and voice cloning
Spark-TTS is an official PyTorch inference implementation of an efficient LLM-based text-to-speech model with single-stream decoupled speech tokens, supporting zero-shot voice cloning and bilingual Chinese/English.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
choose when you need few-shot voice cloning with as little as 1 minute of reference audio and a larger community.
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
choose when you need voice cloning support across 600+ languages rather than only Chinese/English.
High-Quality Voice Cloning TTS for 600+ Languages
choose for a mature open-weight TTS with high expressiveness, if you don't require recent maintenance.
Zonos-v0.1 is a leading open-weight text-to-speech model trained on more than 200k hours of varied multilingual speech, delivering expressiveness and quality on par with—or even surpassing—top TTS providers.
choose for extremely fast inference (150x realtime) on resource-constrained hardware.
A high-quality rapid TTS voice cloning model that reaches speeds of 150x realtime.
choose if an inverted-Whisper architecture fits your multilingual or open-source requirements.
An Open Source text-to-speech system built by inverting Whisper.
choose if you need a broader speech and sound generation model family, not just TTS.
An open-source model family for long-form speech, dialogue synthesis, voice design, sound effects, and real-time streaming TTS
choose for tokenizer-free TTS with creative voice design and true-to-life cloning.
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
choose for industrial-level controllable zero-shot TTS with deployment focus.
An Industrial-Level Controllable and Efficient Zero-Shot Text-To-Speech System
choose for full-stack training/inference support and multi-lingual voice generation.
Multi-lingual large voice generation model, providing inference, training and deployment full-stack ability.
choose for streaming speech generation and free-form voice design from the Qwen ecosystem.
Qwen3-TTS is an open-source series of TTS models developed by the Qwen team at Alibaba Cloud, supporting stable, expressive, and streaming speech generation, free-form voice design, and vivid voice cloning.
choose when you also need zero-shot speech editing, not just TTS.
Zero-Shot Speech Editing and Text-to-Speech in the Wild
choose for controllable TTS covering over 7000 languages.
Controllable and fast Text-to-Speech for over 7000 languages!
choose when emotion/style editing is as important as zero-shot TTS.
A powerful 3B-parameter, LLM-based Reinforcement Learning audio edit model excels at editing emotion, speaking style, and paralinguistics, and features robust zero-shot text-to-speech
choose for cross-lingual zero-shot TTS with broad multilingual coverage.
Confucius4-TTS: a Multilingual and Cross-Lingual Zero-Shot TTS Engine
choose for instant voice cloning with granular control over tone and emotion.
Instant voice cloning by MIT and MyShell. Audio foundation model.
choose for an alternative LLM-based TTS design with its own training pipeline.
An Open-Sourced LLM-empowered Foundation TTS System
choose for emotion-expressive zero-shot TTS trained with multi-reward reinforcement learning.
GLM-TTS: Controllable & Emotion-Expressive Zero-shot TTS with Multi-Reward Reinforcement Learning
choose for a TTS optimized for human-sounding naturalness.
Towards Human-Sounding Speech
choose specifically for Vietnamese TTS with on-device real-time CPU inference.
Vietnamese TTS with instant voice cloning • On-device • Real-time CPU inference • 48kHz audio quality • Chuyển văn bản thành giọng nói tiếng Việt • Text to speech tiếng Việt • TTS tiếng Việt
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
an older expressive TTS model with no recent maintenance.
Foundational model for human-like, expressive TTS
an unmaintained multi-voice prompt-controlled TTS engine.
EmotiVoice 😊: a Multi-Voice and Prompt-Controlled TTS Engine
an archived fast local neural TTS, useful if you need low latency and don't need updates.
A fast, local neural text to speech system
These projects were analysed and named Spark-TTS among their alternatives. The relationship is not symmetric — how Spark-TTS rates them is a separate judgement, made when Spark-TTS is analysed in its own right.
calls Spark-TTS “choose for a simple inference codebase for a competitive TTS model.”
Instant voice cloning by MIT and MyShell. Audio foundation model.
calls Spark-TTS “choose for an Apache-2.0 zero-shot TTS with a compact inference codebase.”
An Industrial-Level Controllable and Efficient Zero-Shot Text-To-Speech System
calls Spark-TTS “Choose when you want a comparable zero-shot TTS with its own inference pipeline and model weights from the Spark project.”
Multi-lingual large voice generation model, providing inference, training and deployment full-stack ability.
calls Spark-TTS “Choose when you want Spark-TTS's specific zero-shot voice cloning architecture.”
Qwen3-TTS is an open-source series of TTS models developed by the Qwen team at Alibaba Cloud, supporting stable, expressive, and streaming speech generation, free-form voice design, and vivid voice cloning.
calls Spark-TTS “Use for a lightweight Apache-2.0 inference-only TTS, but less training support.”
High-Quality Voice Cloning TTS for 600+ Languages
calls Spark-TTS “choose over Zonos if you prefer the SparkAudio model with a straightforward inference codebase; lighter to run than Zonos's full hybrid pipeline.”
Zonos-v0.1 is a leading open-weight text-to-speech model trained on more than 200k hours of varied multilingual speech, delivering expressiveness and quality on par with—or even surpassing—top TTS providers.
calls Spark-TTS “Choose for an actively developed Apache-2.0 TTS with similar cloning/design features; compare quality on your data.”
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
calls Spark-TTS “For a modern TTS inference model with multilingual support, but not optimized for on-device ONNX.”
Lightning-Fast, On-Device, Multilingual TTS — running natively via ONNX.
calls Spark-TTS “Choose when you need a specific Spark-TTS model for local inference; niche spark voice model.”
Use Microsoft Edge's online text-to-speech service from Python WITHOUT needing Microsoft Edge or Windows or an API key
calls Spark-TTS “choose if you want Spark-TTS's specific model and can run GPU inference.”
A TTS that fits in your CPU (and pocket)
calls Spark-TTS “choose for a lightweight zero-shot TTS inference implementation, if you accept sparse updates and no editing.”
Zero-Shot Speech Editing and Text-to-Speech in the Wild
calls Spark-TTS “choose to run Spark-TTS inference, but there's no bundled service layer.”
EmotiVoice 😊: a Multi-Voice and Prompt-Controlled TTS Engine
calls Spark-TTS “Inference code for a recent TTS model; choose it if you specifically want Spark-TTS's voice cloning and language features.”
High-quality multi-lingual text-to-speech library by MyShell.ai. Support English, Spanish, French, Chinese, Japanese and Korean.
calls Spark-TTS “Choose it if you already have a Spark-TTS model and need only inference, not training.”
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
calls Spark-TTS “Narrower bilingual TTS with stale code; only if you need its specific voice capabilities.”
SoTA open-source TTS
calls Spark-TTS “Choose when you need TTS without requiring Apple Silicon/MLX and prefer a lightweight inference-only model.”
A text-to-speech (TTS), speech-to-text (STT) and speech-to-speech (STS) library built on Apple's MLX framework, providing efficient speech analysis on Apple Silicon.
calls Spark-TTS “Older inference-only TTS codebase, superseded and last pushed over 18 months ago.”
A powerful 3B-parameter, LLM-based Reinforcement Learning audio edit model excels at editing emotion, speaking style, and paralinguistics, and features robust zero-shot text-to-speech