21 Text to Speech tools (20 free)
21 Text to Speech resources, picked out of 17 saved Instagram reels and 10 liked tweets, from creators like @itsemilyhiggins, @theaiagents, @benkaluza. 20 of them are free, freemium or open source.
ElevenLabs produces natural, expressive text-to-speech and custom voice clones across dozens of languages. It includes creative audio tools for dubbing, sound effects, and music generation alongside conversational AI agents. Developers and creators can access its voice models through a web interface or API.
Sonic is an AI text-to-speech model built on State Space Model (SSM) architecture, designed for real-time applications and conversational voice agents. It generates expressive speech with sub-90ms latency across 44 languages and can clone voices from just a few seconds of clean reference audio. It also supports inline transcript controls like custom pronunciations and expressive cues like laughter.
Dia is an open-source 1.6-billion parameter voice model developed by Nari Labs that creates conversational audio directly from text transcripts. It supports multi-speaker alternating dialogue alongside nonverbal cues like laughter, sighs, and throat clearing. The model can be conditioned on audio prompts for voice cloning and emotion control, running locally via PyTorch/Transformers or in Hugging Face Spaces.
Airy is a high-speed, budget-friendly text-to-speech platform developed by Out of Set. It delivers speech generation with very low latency (around 200ms) and high throughput at a fraction of standard market costs. The platform provides both a web-based studio interface for content creators and an API for real-time application pipelines.
AudioPen is an AI-powered speech-to-text tool that rewrites and restructures spoken rambling into clear, cohesive drafts rather than just transcribing verbatim. It allows you to select rewrite styles, control output lengths, and dictate directly into apps via system-level hotkeys and mobile keyboards. It is especially useful for quickly turning messy voice memos into finished emails, articles, or notes.
cartesia-python is the official client library for interacting with Cartesia's Sonic text-to-speech models in Python. It supports synchronous and asynchronous HTTP requests alongside WebSocket streaming for low-latency audio generation. It is designed for developers building voice agents and real-time speech applications.
Chatterbox is an open-source text-to-speech model from Resemble AI designed for expressive, zero-shot voice cloning. It can mimic a target voice from just a few seconds of reference audio while providing control over emotional intensity. It runs faster than real time and includes built-in audio watermarking.
Coqui TTS is a widely used open-source library and CLI tool for text-to-speech synthesis and voice cloning. It provides pretrained models (including XTTS) across multiple languages that can run locally on your own hardware. Creators and developers use it as a self-hosted base audio generator for AI avatars, voiceovers, and voice-conversion workflows.
An Apple Shortcuts workflow designed to replace standard device dictation. It records spoken audio, transcribes it through the ElevenLabs Scribe API, refines and formats the text with OpenAI, and copies the clean result directly to the clipboard.
Kyutai TTS is an open-weights, 1.8B-parameter streaming speech model designed for real-time voice applications. It begins generating audio after just a few input words and runs with minimal latency, supporting high-throughput batching on single GPUs. The weights are released under a permissive CC-BY-4.0 license, making it suitable for production conversational agents.
Lovevoice AI is a browser-based text-to-speech platform that converts text and uploaded documents into natural-sounding voiceovers. It provides access to nearly 300 AI voice profiles and over 70 languages with regional accents. It is a practical web alternative to services like ElevenLabs for creating downloadable MP3 narrations without complex setup.
MiniMax Audio is an AI-powered voice and music platform featuring advanced text-to-speech models like Speech 2.8. It allows users to clone voices using just 10 seconds of source audio, design novel voices from text prompts, and generate multi-emotion speech across numerous languages.
This note provides quick benchmark figures and minimal Python code for running OpenAI Whisper models locally on Apple Silicon using the mlx_whisper library. It demonstrates that three hours of audio can be transcribed in under five minutes on high-end chips like M2 Ultra and M3 Max using the whisper-large-v3-turbo checkpoint. It serves as a handy reference for setting up efficient, private speech-to-text without cloud APIs.
Murf AI is an AI voice platform that converts text into natural-sounding voiceovers and real-time speech. It includes a web studio for editing audio with custom pitch, speed, and emphasis, alongside developer APIs for building conversational voice agents. Creators and teams use it to produce narrations, localize video content, and automate voice interactions.
NaturalReader converts text, documents, web pages, and scanned books into spoken audio using natural AI voices. It supports over 90 languages, customizable prompt-based voice styles, and instant voice cloning across web, mobile, and browser extension formats. It caters to both personal reading accessibility and commercial voiceover production.
Parakeet TDT 0.6B V2 is an open automatic speech recognition model from NVIDIA based on FastConformer architecture. It transcribes audio up to 24 minutes in a single pass while outputting punctuation, capitalization, and word-level timestamps. It achieves high benchmark accuracy under a commercially permissive CC-BY-4.0 license.
qwen3_tts_rs is a native Rust implementation for Qwen3 text-to-speech model inference. It packages TTS, voice cloning, and an OpenAI-compatible audio API server into standalone CLI binaries without requiring a Python runtime. It supports hardware acceleration using Apple Silicon MLX via Metal GPU as well as libtorch with NVIDIA CUDA.
Qwen3-TTS-12Hz-1.7B-CustomVoice is an open-source text-to-speech model from Alibaba's Qwen team that allows fine-grained voice style and emotional direction via natural-language prompts. It supports 10 languages and multiple built-in speaker timbres with ultra-low latency streaming synthesis. It is useful for generating realistic dialogue and character voices without robotic cadence.
SPEECHMA is a web-based text-to-speech generator offering over 580 AI voices across dozens of languages and regional accents. It allows users to convert text blocks up to 2,000 characters into natural audio with commercial usage rights. It is worth knowing about as a straightforward, no-signup alternative to paid voice generators like ElevenLabs.
Supertonic is an open-source, on-device text-to-speech model built to execute fast local inference without relying on cloud APIs or dedicated GPUs. Operating via ONNX Runtime with a compact 99M-parameter footprint, it produces 44.1kHz audio across 31 languages on hardware ranging from desktop CPUs to Raspberry Pis. Note that the original repository has been archived, though code and weights remain usable under the MIT license.
Zonos-v0.1 is an open-weight speech generation model developed by Zyphra and trained on over 200,000 hours of multilingual audio. It provides voice cloning from short audio clips alongside conditioning controls for speaking rate, pitch, and specific emotions, delivering audio natively at 44kHz.