1. Which features produce AI speech
Several parts of Vocaic generate synthetic speech:
- Talk, Desk and Studio delivery voices: when a message is delivered as voice, it is spoken by a Vocaic voice in the accent the listener chose.
- Accent Studio re-voicing: your English is re-delivered in a different English accent, using a Vocaic voice.
- The films’ narration and Ava, the AI guide, assistant and room agent: these are AI-generated voices, not recordings of a specific person speaking to you.
- Songs with vocals in Studio: a Vocaic voice is timed and pitched onto a melody in your browser. This is a live preview, and nothing is cloned.
- Agents built with the agent builder: their spoken replies use a Vocaic voice in the language and accent the agent's owner chose.
2. Ava, the AI guide, assistant and room agent
Ava narrates the guided tour, answers questions you type or say on the site, and can join a Talk room as an AI agent. Her spoken voice is a Vocaic voice. Her written answers are generated by an AI model hosted by Vocaic’s cloud provider, Amazon Web Services; your question and a short recent conversation are sent to that model to write the answer, and are not used to train models. A cached answer, without any identity, is kept for 7 days so the same question is faster next time. Ava is AI and can be wrong, so do not share personal details with her and do not rely on her for legal, medical or financial advice. In a Talk room, Ava replies when addressed, and the room delivers her reply to each person in their own language and accent. In a room she can also catch you up, take minutes, list action items and decisions, translate, explain a phrase, rewrite a message, draft a reply, run a poll and set a reminder; polls and reminders are stored with the room and deleted with it. On her page she can create a room, translate and play a phrase, build a phrasebook, summarize text or a public web page link you give her (Vocaic's server fetches that page's text once), draft a message, set a timer, plan a meeting across time zones, convert units and help you practise a language. Notes you ask her to remember stay in your browser and are sent with your questions only as context. In the Ava app, answers appear as they are written, can cite public Wikipedia articles she looked up (a short search query is sent to Wikipedia for that), and can read images and documents you attach; attachments are processed for that answer and not stored, and your conversations stay in your browser. When a business connects Slack, WhatsApp, Telegram or Discord, Ava can answer there too, and everyone receives messages in their own language. On an Alexa skill someone connects, Ava answers out loud in the Vocaic voice and accent the skill owner chose.
3. Delivery voices are Vocaic voices, never cloned
All delivery and narration voices are Vocaic voices generated by Vocaic's speech engine. They are not built from your recordings, and Vocaic does not create a voice clone of you or anyone else. The Accent Studio keeps your words exactly as you said them; it changes only how they are delivered, not their meaning.
4. How transcription works
When you record or upload audio in the Accent Studio, it is transcribed by an open-source speech model running on Vocaic’s own server. The temporary audio upload is deleted after transcription. Built-in live speech recognition uses your browser vendor under that vendor's terms. In browsers without built-in speech recognition, Vocaic recognizes your speech on your device; the audio is not uploaded.
5. How translation works, and its limits
Translation is performed by Amazon Translate (Amazon Web Services) for the 75 languages it covers, and by MyMemory (Translated srl) for the others and as a backup; if both are unavailable, Vocaic's AI model translates the message. Machine translation can be wrong, incomplete or awkward, so the text of every message is always available to read, and you can compare it with the original. Do not rely on a translation for safety-critical, medical or legal decisions (see the Acceptable Use Policy).
6. The recorded accent examples
The Accent Studio includes short recorded English examples drawn from the Edinburgh International Accents of English Corpus (EdAcc), used under the Creative Commons Attribution-ShareAlike 4.0 (CC BY-SA 4.0) licence. These are recordings of real speakers used to illustrate accents, not Vocaic voices.
8. The Studio tools (text to speech, voice design, Voice library, voice changer, isolator, dubbing, editor, podcast, sound effects, music and songs, transcription, ad localizer)
Vocaic Studio includes several tools, each with the accent layer. Every voice is a Vocaic voice or a voice you design by transforming a Vocaic voice (pitch, tone, pace and breath) in your browser; Vocaic never clones anyone's voice. Transcription, script alignment, the voice changer, dubbing, the voice isolator and the editor's captions run Vocaic's on-device speech engine in your browser, so the audio you process stays on your device; the engine's model files are downloaded from Vocaic once and kept in your browser's cache. Text you choose to have spoken is sent to Vocaic's server to be synthesized, and text you translate is sent to the translation service. Sound effects and instrumental music are synthesized in your browser from a recipe or a score. Songs with vocals are sung by a Vocaic voice that your browser times and pitches onto the melody, a live preview. The Voice library shares designed voices only: a Vocaic voice plus design settings, never a recording. Match my settings listens to your microphone for a few seconds to measure pitch, pace and brightness on your device; nothing is recorded or kept, and it does not copy your voice. Audio exports as WAV or MP3, encoded on your device. Podcast drafts, voice design suggestions, sound recipes, music scores, song lyrics and melodies, and localized ad copy are written by Vocaic's AI, hosted by Vocaic's cloud provider, Amazon Web Services, and are not used to train models. Dubbed or re-voiced media must respect the rights of the people in it and the acceptable use policy.
9. Questions
Questions about how AI and voice work here can be sent to agarw574@purdue.edu.