You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 00d4e7c
Browse filesBrowse the repository at this point in the historyBrowse files
When extraction runs on an OpenAI reasoning model (the GPT-5 family, GPT-5.6, or GPT-6), set `model.reasoningEffort` to control how much the model reasons before it extracts. Lower it to save cost and latency, or raise it for harder extraction prompts.
616
+
617
+
Each model accepts only the efforts OpenAI supports for it, for example `none` and `xhigh` on GPT-5.6, or `minimal` on GPT-5. Setting an effort on a non-reasoning model such as GPT-4.1, or on a non-OpenAI model, returns a `400`. When unset, extraction uses the model's current default. Reasoning effort applies to the primary extraction only; the GPT-4.1 fallback runs without it. Reasoning tokens are billed as output tokens.
1.**GPT-Live voices**: Nine new [GPT-Live](/gpt-live/configuration) voices are available: Nuri, Juni, and Flitz (German), Brise and Sillage (French), Nira (Hindi), Harema and Shida (Japanese), and Noeul (Korean). Assistants that don't use a GPT-Live model can no longer be saved with a GPT-Live only voice.
6
+
7
+
## Features
8
+
9
+
1.**GPT-Live prompt editor**: The [GPT-Live](/gpt-live/configuration) speaker and reasoner prompts now use the rich text editor, with visual and code views and a fullscreen mode.
10
+
2.**Reasoning effort for structured outputs**: You can now set a reasoning effort for [structured outputs](/assistants/structured-outputs#reasoning-effort) that use an OpenAI reasoning model.
11
+
3.**Deepgram feedback filter setting**: Set `feedbackFilterEnabled` to false on a [Deepgram](/providers/transcriber/deepgram#feedback-filter) transcriber to keep caller speech that repeats what the assistant just said. It's on by default.
12
+
4.**Simulation suite generation API**: You can now draft requirements and generate [simulation suites through the API](/observability/simulations-generate-api) with a private API key.
13
+
14
+
## Fixes and Improvements
15
+
16
+
1.**ElevenLabs pronunciation dictionaries**: Calls that use an [ElevenLabs](/providers/voice/elevenlabs) pronunciation dictionary without a pinned version no longer fail. They use the dictionary's latest version.
17
+
2.**ElevenLabs silent turns**: When ElevenLabs returns no audio for a turn, the call retries once and then switches to your [fallback voice](/voice-fallback-plan).
18
+
3.**GPT-Live phone audio**: [GPT-Live](/gpt-live/configuration) phone calls no longer have short gaps of silence in the middle of words.
Copy file name to clipboardExpand all lines: fern/customization/speech-configuration.mdx
+1-3Lines changed: 1 addition & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -10,9 +10,7 @@ description: Control when a Vapi assistant starts and stops speaking, tuning spe
10
10
Speech configuration lets you control exactly when your assistant starts and stops speaking during a conversation. By tuning these settings, you can make your assistant feel more natural, avoid interrupting the customer, and reduce awkward pauses.
11
11
12
12
<Note>
13
-
Speech speed can be controlled, but only PlayHT currently supports this
14
-
feature with the `speed` field. Other providers do not currently support
15
-
speed.
13
+
Some voice providers support a `speed` field to control how fast the assistant speaks. Support and the accepted range vary by provider.
Copy file name to clipboardExpand all lines: fern/gpt-live/configuration.mdx
+18-2Lines changed: 18 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -20,7 +20,7 @@ GPT-Live supports a different set of features from Vapi's other voice architectu
20
20
| Browser WebRTC, native Twilio, and Vapi SIP | Supported |
21
21
| Raw WebSocket | Supported with mono, signed 16-bit little-endian PCM at 24 kHz. Set the rate explicitly |
22
22
| Native Telnyx or Vonage | Not supported |
23
-
|22 preset OpenAI voices | Supported. See [voices](#voices)|
23
+
|31 preset OpenAI voices | Supported. See [voices](#voices)|
24
24
| Custom or cloned voices, or other voice providers | Not supported |
25
25
| Changing voice during a call | Not supported. Save the assistant and start a new call |
26
26
| Model and voice fallbacks | Not supported. GPT-Live also can't be a fallback model |
@@ -161,7 +161,7 @@ Keep delegation triggers in the prompt in the same language. Try names, dates, a
161
161
162
162
## Voices
163
163
164
-
Choose any of these 22 voices with `voice.provider: "openai"` and `voice.voiceId`. Save the assistant and start a new call to change voices. In-call voice changes aren't supported.
164
+
Choose any of these 31 voices with `voice.provider: "openai"` and `voice.voiceId`. Save the assistant and start a new call to change voices. In-call voice changes aren't supported.
165
165
166
166
### Voices with regional tags
167
167
@@ -197,6 +197,22 @@ Use the tags to narrow your choices, then listen to the samples. Bossa and Tempo
197
197
| Shimmer |`shimmer`|[Listen to Shimmer](https://voices-storage.vapi.ai/gpt-live/marble-alpha-v6/shimmer.mp3)|
198
198
| Verse |`verse`|[Listen to Verse](https://voices-storage.vapi.ai/gpt-live/marble-alpha-v6/verse.mp3)|
0 commit comments