Skip to content

Transcription and Text Threads

Conversation Analytics needs text to analyze. For voice calls this comes from transcription; for text channels it comes from the message/thread text.

Voice calls: transcription

For calls, MiaRec converts audio into a transcript:

  • a time-ordered text representation of what was said
  • often including speaker turns (agent vs customer)
  • sometimes including timestamps (depending on configuration)

Why transcription quality matters

AI insights depend on what the model “sees” in the transcript. Common factors that influence quality include:

  • background noise or poor audio
  • overlapping speech
  • accents and domain-specific vocabulary
  • incorrect speaker attribution

If the transcript is incomplete or inaccurate, insight quality may degrade.

Multi-lingual support

MiaRec supports conversations in multiple languages:

  • Transcription engines support multiple languages and can auto-detect the spoken language
  • AI models (LLMs) are multilingual and can analyze transcripts in their original language
  • Optional translation to English is available for reviewers who prefer to read transcripts in English

When the source language differs from the reviewer's preferred language, the system can translate both the transcript and the AI-generated insights (such as summaries and explanations).

Text channels: threads and messages

For chats, emails, and tickets (when enabled), MiaRec analyzes the text directly:

  • chat messages in order
  • email thread content
  • ticket messages/notes

For accurate analysis, it helps to preserve:

  • who said what (agent vs customer)
  • message ordering and timestamps
  • relevant metadata (queue, tags, status)

Minimum content requirements

Some conversations may be too short or lack enough evidence for certain insights. In those cases, MiaRec may:

  • skip running a task (if filters exclude it)
  • produce “unknown / not enough evidence”
  • produce partial insights (depending on the task)

Next: how results are stored

Once AI extracts insights from text, results are stored in structured fields: