Modern TTS solved intelligibility years ago. The Blizzard Challenge, an annual benchmark that has tracked speech synthesis progress since 2005, reported that by 2021 synthetic speech had become indistinguishable from natural speech in intelligibility, and effectively indistinguishable in naturalness too (Perrotin et al., 2024). So the field moved on. The interesting question stopped being can you understand the voice and became does the voice sound like it means what it's saying. Speechify Studio answers it with emotional text to speech, on top of plain text to speech. For the developer API, the answer depends on the model, and the current models do not apply emotion styles.
Emotion Styles Are a Simba 1.6 Feature
On the API, emotion styles come from the <speechify:style> SSML tag, and only the Simba 1.6 models, simba-english and simba-multilingual, apply it. Its thirteen values are Angry, Cheerful, Sad, Terrified, Relaxed, Fearful, Surprised, Calm, Assertive, Energetic, Warm, Direct, and Bright. Simba 1.6 is retired for new workspaces, and from 21 November 2026 both of its ids are served by the current models, which accept the tag without applying it. See Migrating off Simba 1.6.
Using Emotions in the Speechify AI Voice Generator
The AI Voice Generator lives inside Speechify Studio and is the tool creators use for voiceovers, marketing videos, audiobooks, and podcast segments. You paste your script, pick a voice, and then apply an emotional style to the whole clip or to a specific selection. Because emotions are per-selection, the same voiceover can carry a Cheerful hook, an Assertive value proposition, and a Warm sign-off without switching voices.
A few concrete workflows where this matters: marketing videos rendered inside a tool like CapCut usually need two or three tonal beats, such as attention, promise, call to action. Instead of three separate voice takes, you generate one voiceover with the emotion changing per line. Audiobooks and fiction narration benefit even more, since dialogue between characters can carry different emotional weight without hiring multiple readers. For explainers, a single Calm voice reading a dense middle section reads clearer than the same voice attempting to sound "engaging" the whole way through.
What the Speechify API Applies Today
On the current models, simba-3.2 and simba-3.0, the Speechify API accepts <speechify:style> and returns 200, but speaks the text inside the tag without the style. The same goes for <prosody pitch>, <prosody volume> and <emphasis>. Three SSML tags do apply: <break> for pauses, <prosody rate> for speaking rate, and <sub alias> for pronunciation. The SSML guide lists what each model applies.
For example, <speak>Your order has shipped. <break time="500ms" /> <prosody rate="slow">Read the <sub alias="Frequently Asked Questions">FAQ</sub> for delivery times.</prosody></speak> pauses for half a second, then reads the second sentence slowly and says "Frequently Asked Questions" in place of FAQ.
Delivery on the current models comes from the text and the voice. Punctuation and sentence shape set the phrasing, so write each line the way you want it said, and add a <break> where you need a longer pause than punctuation gives. The voice you choose sets the baseline character. For e-learning feedback, that can mean a short, plain line for a correct answer, <prosody rate="slow"> for a hint, and a <break> before the explanation.
Speechify AI Voice Generator vs Speechify API: Which One to Pick
Where Emotional Voices Actually Change What You Can Make
Emotional TTS is the missing piece for anything creative. A single celebrity-style voice reading a full audiobook, an animated character delivering both punchlines and grief, a podcast intro that lands on the right beat — none of these worked with flat synthesis. They work now because the emotion sits inside the voice, not in the script direction. For creators already using Speechify's celebrity and character voices, emotional styles are the difference between a voice that sounds like the reference and a voice that acts like it.
Does Emotional Delivery Actually Improve Comprehension?
Yes, and this is measurable outside the AI world too. In a 2022 study of 11 to 13-year-olds and a 2025 replication, researchers Dylman and Champoux-Larsson found that listeners answered more content questions correctly when the same factual material was read with positive emotional prosody than with a neutral tone (Dylman et al., 2025). The comprehension gain was not small, and it held across immediate and delayed recall. For educational content, product tutorials, and any long-form audio where retention matters, an emotionally appropriate voice is a genuine comprehension aid.
FAQ
What is text to speech with emotions?
It is synthetic speech that carries a chosen emotional tone, like cheerful or sad, on top of the words, so the same sentence can be delivered multiple ways. In Speechify Studio's AI Voice Generator you pick the emotion per selection. The Speechify API's current models do not apply emotion styles.
How many emotions does the Speechify API support?
On the current models, none: simba-3.2 and simba-3.0 accept the <speechify:style> tag and speak the text without the style. Its thirteen values, from Angry to Bright, are applied only by the retired Simba 1.6 models. The AI Voice Generator in Speechify Studio has its own emotion selector, described above.
Where do I control emotion in the AI Voice Generator?
Inside Speechify Studio, after you enter your script, an emotion selector appears next to the voice picker. Apply it to the whole clip or to a highlighted selection, and re-render.
Can I set an emotion in the Speechify API?
Not on the current models. simba-3.2 and simba-3.0 accept <speechify:style emotion="..."> and return 200, but speak the text without the style. Shape delivery through the text and the voice you choose, and use the tags they apply: <break>, <prosody rate> and <sub alias>. See the SSML guide.
Can I combine emotions in a single voiceover?
In Speechify Studio, emotions apply per selection, so one voiceover can shift tone line by line. On the API's current models emotion styles are not applied. What you can vary per span is speaking rate with <prosody rate> and pauses with <break>.
Do emotional voices sound as natural as neutral ones?
Very close. Peer-reviewed emotional TTS models now score around 3.94/5.0 for expressiveness and 3.98/5.0 for naturalness, meaning listeners rate them almost equally on both dimensions.
Does emotional delivery actually help listeners retain the content?
Research on human speakers says yes. Positive prosody improved recall of factual content in controlled studies. The same principle applies to well-directed AI voiceovers.
Can I use emotional voices for commercial voiceovers?
Speechify licensing covers commercial use of generated voiceovers, including emotional styles. Check your plan for output-length limits.
How is this different from just adjusting speed or pitch?
On Simba 1.6, emotion styles change prosody as a whole, including pauses, stress, intonation contour, and energy, which is why an "angry" version does not sound like a fast, high-pitched version of the neutral one. On the current models, speed is the part you can adjust: <prosody rate> applies, while <prosody pitch> and emotion styles are accepted but not applied.
