Leaving Hume EVI
Hume announced that API access to EVI and its text-to-speech ended on November 13, 2026 at 12:01 a.m. EST. EVI bundled transcription, a language model, a voice and expression scores, so replacing it means choosing each part. Oruk covers two of them: transcription, and scores for how each phrase sounded. It does not generate speech.
Updated
The announced EVI and TTS cutoff has passed
The EVI/TTS notice reviewed October 3, 2026 announced that API access cutoff was November 13, 2026 at 12:01 a.m. EST (November 12, 2026 at 9:01 p.m. PST). Do not rely on Hume as a migration fallback.
The historical notice states that account data is permanently deleted after November 13, 2026. Use your own verified archive; do not assume Hume can recover missing data.
Sources reviewed . Oruk is not affiliated with Hume AI.
What replaces each part
| EVI did this | Replace it with | From Oruk? |
|---|---|---|
| Transcribed the user | Oruk Realtime: live words in 32 locales | Yes |
| Scored vocal expression | Oruk Realtime phrase emotions, or Resonance 2 for scores without a transcript | Yes |
| Chose what to say and called tools | A language model you pick | No |
| Spoke the reply | A speech synthesis provider you pick | No |
| Handled turns and interruptions | An agent framework such as Pipecat or LiveKit Agents | No |
| Carried the audio | Your WebRTC or telephony provider | No |
The scores are different
| Oruk | |
|---|---|
| Labels | 15 emotions and 16 speaking styles |
| Emotions | happy, excited, hopeful, sad, worried, angry, frustrated, disappointed, scared, disgusted, surprised, embarrassed, proud, relieved, neutral |
| Styles | energetic, passionate, irritated, warm, playful, sarcastic, deadpan, hesitant, confident, sincere, skeptical, tired, formal, casual, impatient, distracted |
| Scale | Each score runs from 0 to 1 on its own. Scores do not sum to one. |
| Timing in Realtime | One result for each phrase, in an event after the transcript |
| Compatibility | No mapping to Hume’s labels or thresholds is provided. Recalibrate on your recordings. |
The Oruk side of the stack
import WebSocket from "ws"
const socket = new WebSocket(
"wss://speech-api.oruk.ai/v1/realtime?model=oruk-realtime",
"oruk-realtime",
{ headers: { Authorization: `Bearer ${process.env.ORUK_API_KEY}` } },
)
socket.on("open", () => socket.send(JSON.stringify({
type: "session.update",
session: { model: "oruk-realtime", language: "auto", sample_rate: 16000, phrase_emotions: true },
})))
socket.on("message", (data) => {
const event = JSON.parse(data.toString())
if (event.type === "conversation.item.input_audio_transcription.completed") onTranscript(event)
if (event.type === "conversation.item.input_audio_emotion.completed") onTone(event) // join by phrase_id
})Send mono 16-bit PCM as binary frames. Pass each transcript, with its tone scores, to your language model.
In order
- Export what you need from Hume and check the archive opens.
- List what your application reads from EVI: transcripts, specific expression scores, tool calls, voice.
- Choose a framework, a language model and a voice. Port prompts and tool definitions.
- Add Oruk Realtime for transcripts and tone. Set any thresholds again on your recordings.
- Test interruptions, tool failures and reconnection on real calls before you cut over.
Questions
- Is Oruk a drop-in replacement for EVI?
- No. Oruk does not run the conversation, call tools or speak. It replaces the transcription and the expression scores. You choose the language model, the voice and the transport.
- Do Oruk scores match Hume’s?
- No. Oruk returns 15 emotion and 16 speaking-style labels with their own scale. Thresholds tuned on Hume scores need to be set again on your own recordings.
- Is Hume’s Expression Measurement API also ending?
- Hume’s notice covers EVI and text-to-speech. It does not say the Expression Measurement APIs are ending. Confirm current access and terms with Hume.
- Can I keep my Hume voice?
- Not through Oruk. Voice IDs do not transfer between providers. Check your rights to any custom voice with the speech synthesis provider you move to.
- What should I export from Hume?
- Chat history, configurations, prompts, tool definitions and any recordings you are permitted to keep. Hume’s notice says account data is permanently deleted after November 13, 2026.
Try it on your own audio.
The demo needs no account. The trial needs a card.