STT Filler Words
# supportm
mud
05/02/2025, 1:12 AMHi,
I'm looking to get filler words from the user speech - in the transcript - with the idea to determine hesitation / confidence.
Is that something that is intentionally removed from STT in the transcript?
v
Vapi Support Bot
05/02/2025, 1:13 AMBased on the available information, I cannot definitively state whether filler words are intentionally removed from STT transcripts by default. However, Vapi does recognize the importance of filler words like "umm," "ahh," and "like" in conversational speech. For more detailed control over transcription settings, you would need to refer to the documentation of your specific transcription service provider (such as Deepgram or AssemblyAI).
Sources:
- [Vapi Glossary](https://docs.vapi.ai/glossary)
- [How Vapi Works](https://docs.vapi.ai/how-vapi-works)
- [Discord Channel](https://discord.com/channels/1211482211119796234/1303981615344779325)
k
Kings_big💫
05/02/2025, 2:54 PMVapi Talkscriber transcription service includes filler words like “uh” and “um” to preserve natural speech flow, which can help analyze hesitation and confidence.
m
mud
05/02/2025, 9:31 PMwhat do you mean by 'Vapi Talkscriber transcription service'?
is it the default service, or an additional service?
k
Kings_big💫
05/03/2025, 9:25 AMThe Vapi Talkscriber transcription service is an optional built-in provider offering accurate, real-time speech-to-text and emotion detection, not the default but selectable in settings.
m
mud
05/03/2025, 8:10 PMfound it, thanks
what do you mean by emotion detection in terms of transcription?
does it indicate any emotion in the transcript?
mud
05/03/2025, 8:17 PMhow do i tailor my prompt to make use of the emotion detection, and how can i get info about the emotion after the call?
k
Kings_big💫
05/03/2025, 9:17 PMLike tone and pitch to identify the speaker’s emotions
Kings_big💫
05/03/2025, 9:17 PMTailor your structuredDataPrompt and structuredDataSchema to include sentiment or emotional cues, and retrieve the post-call emotion data from call.analysis.structuredData via the Vapi API.
m
mud
05/03/2025, 10:04 PMhmm,
but these - structuredDataPrompt and structuredDataSchema
- only work on the transcript right?
if the transcript doesn't have details of tone / pitch, then how would it use this?
mud
05/04/2025, 6:55 AMI want to get tone and pitch/ emotion from the speaker
I was wondering what you use to get the tone and pitch from?
Is it only derived from the transcript?
k
Kings_big💫
05/04/2025, 9:15 AMYes, structuredDataPrompt and structuredDataSchema work on the transcript, so if the transcript doesn’t include tone or pitch details, those elements won’t be available unless you integrate additional tools to analyze audio features..
Kings_big💫
05/04/2025, 9:17 AMVapi detects tone, pitch, and emotion through audio analysis, not just the transcript, using techniques like pitch variation detection and integrating with tools like Talkscriber for emotion detection during conversations..
m
mud
05/05/2025, 4:58 AMRegarding this part -
since vapi detects tone, pitch, emotion etc - can it provide this information in the metadata of the transcript?
It would be for us to useful to analyze the call using this information
k
Kings_big💫
05/05/2025, 5:59 AMYes, Vapi can include emotion, tone, and pitch data in transcript metadata by configuring the analysisPlan with a custom structuredDataSchema, retrievable via the API after the call..
m
mud
05/06/2025, 8:50 PMwill look into this
thanks for the help!