Skip to main content
DialNexa transcripts, recordings, and summaries help users reconstruct what happened in a call. The recording captures sound, the realtime transcript shows what the agent likely heard during the live conversation, the post-call transcript supports cleaner review where available, and the summary compresses the conversation.
Use the transcript for words, the recording for reality, and the summary for speed. Mixing those jobs creates confusion.

Evidence Type Comparison

EvidenceBest forLimit
RecordingAudio quality, tone, interruptions, pauses, pronunciation, and whether the caller actually said something.Takes time to review.
Realtime transcriptUnderstanding what Deepgram or Soniox produced while the agent was deciding what to say.May include recognition errors caused by noise, accents, endpointing, or overlap.
Post-call transcriptCleaner after-call review where available.May not explain the exact live timing that caused an interruption or delay.
SummaryFast operational understanding.Not a verbatim record and should not be treated as legal-grade call text.

Realtime Versus Post-Call Transcript

QuestionUse realtime transcriptUse post-call transcript
Why did the agent reply too early?Yes, because live boundaries matter.Helpful, but not enough on its own.
Did the caller provide a fact?Yes, if the agent acted on it live.Yes, for after-call review.
Why is post-call extraction wrong?Useful for raw evidence.Often the better text to compare with extraction.
Did audio noise affect recognition?Use with recording.Use with recording.

Review Speech Quality

1

Start with recording

Listen for noise, overlap, silence, pace, pronunciation, and whether the call path sounds clipped.
2

Read realtime transcript

Check what text the model likely used in the moment.
3

Compare post-call transcript

See whether after-call cleanup changes interpretation.
4

Use summary last

Confirm the summary matches the evidence, not the other way around.
5

Trace the issue to the right owner

If the text is wrong, tune transcriber or audio. If the text is right but the reply is wrong, tune prompt, model, functions, or variables.

Transcript And Recording Issues

The call may not have a recording URL, recording may still be processing, or sensitive data may have been deleted.
The transcript may not exist for that call or may have been removed.
That can happen because live transcription optimizes for speed, while post-call processing can be cleaner. Use realtime for live behavior and post-call transcript for review.
Improve post-call field definitions, call outcome instructions, or transcript quality before assuming the summary system is broken.
The copy action filters unhelpful tool-call entries where possible, but review call context before sharing transcripts externally.

Speech To Text

Compare Deepgram, Soniox, and transcript behavior.

Audio Quality

Diagnose noisy calls.

Call Detail Page

Use the full detail view.

Annotations

Mark bad turns.