Real-time text to speech and voice cloning from our Simba speech models. One API, 500K characters free every month. Devs, start here → https://xitter.cf/t.co/7LA24XTTPT
Worldwide
Joined March 2017
- Tweets1.4K
- Following42
- Followers4.1K
- Likes816
Pinned Tweet
What's your way of showing you care?
In Seoul, it can sound like "Have you eaten?"
In Moscow, like "Put your hat on, it's minus twenty out."
In Tbilisi, like a toast to the guests.
We built Simba 3.5 to connect people in 48 languages, spoken like a local. Out today.
Sound on.
🤖 Made with AI
SpeechifyAI retweeted
Meet Simba 3.5: one model, 48 languages, streaming-native. Compatible with every voice, including cloned voices.
Congrats to the @SpeechifyAI team, we're proud to be your partner for inference!
For developers: one model id, simba-3.5. Set the language on the request, and every voice in the catalog speaks all 48, including one you cloned.
Launch price: half off until 21 Nov.
The request and the full list: speechify.ai/blog/simba-3-5?…
🤖 Made with AI
In blind side-by-side listening, English listeners preferred Simba 3.5 over Simba 3.2 63% of the time.
The fact sheet: speechify.ai/reports/simba-3…
🤖 Made with AI
For developers: one model id, simba-3.5. Set the language on the request, and every voice in the catalog speaks all 48, including one you cloned.
Launch price: half off until 21 Nov.
The request and the full list: speechify.ai/blog/simba-3-5?…
🤖 Made with AI
In blind side-by-side listening, English listeners preferred Simba 3.5 over Simba 3.2 63% of the time.
The fact sheet: speechify.ai/reports/simba-3…
🤖 Made with AI
SpeechifyAI now takes text over a WebSocket as your LLM writes it, and speaks each sentence the moment it is complete. Measured from US East on 9 Oct 2026, the first word was audible 49 ms after the LLM finished its first sentence.
No sentence splitter or request queue on your side. How it works, and every session we timed: speechify.ai/blog/websocket-…
We timed 13 text-to-speech endpoints from US East on 9 Oct 2026 and published every request.
Simba 3.2 had the lowest first audible audio of the 13: a median of 47 ms over our WebSocket, 64 ms over HTTP.
The script and every request: speechify.ai/blog/choosing-a…
Our Text-to-Speech API can now limit or block one end user of your app, not your whole account.
Send safety_identifier, a hash of your own user id, and ask us to cap or block that user. They get a 429 or a 403; your other users never notice.
ALT Limit one end user. Not your whole workspace. A row of ten dots with one hollow, above the line safety_identifier on the SpeechifyAI text-to-speech API, beside a silver wave sculpture.
On 13 Oct 2026, none of the seven other TTS APIs we checked documented a way to do this. How it works: speechify.ai/blog/tts-api-pe…
The full table, the arithmetic for every cell and the sources: speechify.ai/blog/tts-api-pr…
SpeechifyAI retweeted
SpeechifyAI's Simba 3.2 is now the fastest model on the @voicearena_ai US English 🇺🇸 latency board: 123.0 ms to first audio at the median, climbing from #9 at 392.5 ms in its earlier series.
SpeechifyAI shipped serving-side updates, our latest measurement series picked them up, and the ranking moved. That is exactly what measuring in public is for.
At 123.0 ms it sits 45.5 ms ahead of Inworld TTS 2 at 168.5, more than three times faster than its own earlier number, and with a quality rank range of 6 to 8 on the same 20-model board it joins the quality-versus-speed frontier.
Rank the board by TTFA yourself → voicearena.com/tts-leaderboa…
Congratulations to @SpeechifyAI 👏
SpeechifyAI retweeted
Two independent benchmarks in two days, both put Simba 3.2 at blazing fast speeds, while being top for benchmark quality and 10x more affordable than labs.
Coval measured Simba 3.2 at 106 ms to first audio, Voice Arena at 123 ms, the fastest on their US English board. Same voices, same price, no code change. Chart: Voice Arena
SpeechifyAI's Simba 3.2 is now the fastest model on the @voicearena_ai US English 🇺🇸 latency board: 123.0 ms to first audio at the median, climbing from #9 at 392.5 ms in its earlier series.
SpeechifyAI shipped serving-side updates, our latest measurement series picked them up, and the ranking moved. That is exactly what measuring in public is for.
At 123.0 ms it sits 45.5 ms ahead of Inworld TTS 2 at 168.5, more than three times faster than its own earlier number, and with a quality rank range of 6 to 8 on the same 20-model board it joins the quality-versus-speed frontier.
Rank the board by TTFA yourself → voicearena.com/tts-leaderboa…
Congratulations to @SpeechifyAI 👏
Two independent benchmarks in two days: @covaldev measured Simba 3.2 at 106 ms to first audio, @voicearena_ai at 123 ms, the fastest on their US English board.
Same voices, same price, no code change. Chart: Voice Arena
xitter.cf/voicearena_ai/status/2…
ALT Voice Arena Quality vs Speed chart, 1 day, p50: SpeechifyAI Simba 3.2 is the leftmost point in the most attractive quadrant, on the Pareto frontier with Inworld Realtime TTS 2 and Cartesia Sonic-3.6.
SpeechifyAI's Simba 3.2 is now the fastest model on the @voicearena_ai US English 🇺🇸 latency board: 123.0 ms to first audio at the median, climbing from #9 at 392.5 ms in its earlier series.
SpeechifyAI shipped serving-side updates, our latest measurement series picked them up, and the ranking moved. That is exactly what measuring in public is for.
At 123.0 ms it sits 45.5 ms ahead of Inworld TTS 2 at 168.5, more than three times faster than its own earlier number, and with a quality rank range of 6 to 8 on the same 20-model board it joins the quality-versus-speed frontier.
Rank the board by TTFA yourself → voicearena.com/tts-leaderboa…
Congratulations to @SpeechifyAI 👏