Skip to main content

​
Play, search, and grade recordings from the Recording Library

The Recording Library at Quality → Recordings is the global playback hub: every recording your tenant captures — voice calls, conferences, browser softphone calls, video rooms, and digital interactions — joined to its diarised transcript, its latest linked QA scorecard score, and its recording-quality (QC) verdict in one searchable list. It is the supervisor’s QA worklist. The filters below are the direct answer to questions like “show me recordings where the customer said X, scored below Y, or QC-failed” without paging through call rows one by one. Access is restricted to owners, admins, and supervisors — the library surfaces transcript content and QA scores across all agents, so other roles see an access notice instead of the list.

​
What the hub aggregates

Each row of the list presents one recording with:
  • Classification + channel — an icon and label telling voice/sbc, AI agent, browser softphone, conference, video room, or digital chat/email/social/messaging apart, with the call duration beside it.
  • Timestamp — the recording’s start time rendered in your timezone.
  • QC verdict — a QC failed badge when the recording-health check rejected the captured media (for example undecodable or empty audio), so playback trouble surfaces as a list verdict instead of a broken player.
  • QA linkage — the score of the latest linked evaluation, an AI-scored badge when that evaluation was machine-authored, and a flagged badge when it sits below your auto-score flag threshold.
  • Transcript status — transcribed, pending, failed, or skipped; when your full-text search hits, the matched fragment appears inline with its timestamped offset.

​
Filter the library

Three controls at the top of the page narrow the hit list; they combine:
  • Transcript search — free text matched against finalized, diarised transcript fragments (“What did the customer say?”). The query debounces on pause and hits fragments rather than whole transcripts, and the first match renders inline on each row with its offset timestamp.
  • AI-scored only — restrict to recordings whose latest linked evaluation was auto-scored by the AI.
  • Has a QA evaluation — restrict to recordings already linked to any evaluation (auto- or reviewer-authored).
Every filter is also available over the API as query parameters on GET /api/v1/quality/recording-library, which additionally accepts classification, transcription_status, qc_status, a min_score/max_score band on the linked evaluation’s total score, and a since/until time window — so a weekly review batch (for example “support queue, AI-scored, score under 70”) is reproducible from a saved URL. The list paginates with a stable cursor; Load more extends the rendered window without a page refresh.

​
Play with scrubbing

Open a recording from the list’s Open call link — it lands on the call detail page where the replay panel mounts an audio player:
  • Click-to-seek — click anywhere on the waveform band to seek; the playhead and the transcript highlight move together.
  • Skip back / forward 10 seconds — dedicated controls for repeated supervisor rewinds when a phrase needs a second listen.
  • Transcript-to-audio sync — click any transcript line to seek the audio to that turn’s start offset; the player derives per-turn offsets from the recording duration even when diarised timestamps are stored, and only mounts the audio element once media is available.
If the signed playback URL has expired mid-listen (the QC layer reports media errors), the panel surfaces an inline error instead of a silent stall.

​
Drill into the transcript

The call detail page carries the diarised transcript — each speaker turn runs against its own offset for readable parameter-free quoting. For a caption-grade export, the recording’s transcript is available as WebVTT/SRT per recording. Transcription status on each library row tells you before you open it whether the text layer is finalized; search skips non-finalized text rather than returning partial matches.

​
Grade with the scorecard while you listen

The right pane of the call detail page renders the supervisor-only QA scorecard — the same rubric a reviewer scores under Quality → Evaluations. You play, scrub, and grade in one surface: dispatch a score from the scorecard panel while the replay panel and transcript stay open. A score submitted while the audio still plays attaches to that recording joins it immediately in the library — the Has a QA evaluation filter flips it into the reviewed band and the score column updates. Grading is reviewer-scope: the scorecard pane appears for owners, admins, and supervisors only; agents see their own evaluations and the acknowledge/appeal lifecycle instead. Recording capture, retention, and disclosure are tenant-owned controls:
  • Per-call recording is an organizational decision. Enable or disable capture per call or per named recording configuration; the QC layer reports the verdict either way, but no recording is captured unless the tenant turned capture on.
  • Retention is a tenant-tuned window. A named recording configuration authors a retention_days window between 7 and 3650 days per preset. Recordings older than the window clean up according to that tenant control.
  • Disclosure remains the tenant’s responsibility. Whether playing back a recording, quoting a transcript fragment, or exporting a caption file, the reviewer operates inside your organizational access policy; the supervisor/admin access restriction on this page is the dashboard-side guard for that.
Because recording and transcript access stays within your own tenant surface and the retention window is a per-tenant preset, the library itself adds no new data-handling surface — it exposes inside the dashboard what the unified recordings table already holds for your tenant.

​
Common use cases

  • Targeted QA coaching — filter to AI-scored below your flag threshold, open the flagged rows, and let the score play while you read the flagged transcript offsets.
  • QC failure triage — filter to the QC failed badge rows to isolate recordings whose capture broke before they consume review time.
  • Phrase audit — use the transcript search box for a compliance phrase (“cancel”, “refund”) and work the hit list as the review queue.

​
See also