Skip to content

Projects & Recordings

Every recording lives in a project on the Conversations page. This page is the reference for what all projects have in common: how they are created and configured, what you can upload, how transcription progresses, and what to do when it fails.

For what to do with the transcripts afterwards, see the Conversation Analysis Guide (interviews and focus groups) or the Call Center Analysis Guide (call recordings).


  1. Click Conversations in the left navigation
  2. Click New Project
  3. Answer “What type of audio files will you be analyzing?”
  4. Click Next, enter a Project Name and click Create

Choosing the audio type in the New Project wizard

Audio typeIn the wizardUse it for
InterviewsOne-on-one interview recordingsInterviews with one participant; analyzed with the Conversation Analysis Guide
Focus GroupsGroup discussion recordings with multiple speakersGroup discussions; adds the Per Speaker analysis scope
Call Center CallsCustomer service or sales call recordingsContact center calls; unlocks COPC Evaluation, Calls Analysis, call metadata, automatic analysis and the Call Center Report — see the Call Center Analysis Guide

Open the project and click Configuration. The Project Settings dialog has two parts: “Transcription settings cannot be changed after a project is created.”

These are greyed out, so you can check what the project was created with:

  • Audio File Type — the audio type chosen in the wizard
  • Language — always Auto: the language is detected automatically, including recordings that mix languages
  • Return Timestamps — always on. Timestamps are what let you jump from a quote in an analysis result to the moment in the transcript

Edit these whenever you like and click Save.

Delete audio recordings after (days) — recordings are erased permanently this many days after upload. Transcripts and analysis results are kept, so past work stays readable after the audio is gone. Leave the field empty to follow your team’s retention policy.

Transcribe recordings on private infrastructure only — recordings are processed by a speech recognition service operated by Uspeech and never sent to a third-party provider.

  • Transcription may take longer
  • A recording that cannot be processed privately fails rather than being sent outside
  • Your team can require this for every project, in which case the checkbox is on and locked, with a note explaining why

Number of speakers — appears once private transcription is on. It tells the recognizer how many people are on the recording instead of leaving it to work that out.

OptionWhen to use it
Detect automaticallyThe number varies between recordings — interviews with a second participant, focus groups, calls that get transferred
1–10 speakersEvery recording in the project has the same number of voices

Stating the count is both faster and more reliable: left to itself, the recognizer can split one voice across two speakers. Call projects default to 2 speakers.

Private transcription and the speaker count apply to new transcriptions only — recordings already transcribed are unaffected.

Automatic analysis (call projects only) — Call Center Calls projects also carry Analyze calls automatically as they arrive. See Automatic Analysis.


Open the project’s Files tab and drag files onto the upload area, or click it to pick them. Transcription starts by itself — there is no button to press after uploading.

KindFormatsWhat happens
AudioMP3, WAV, M4A, AAC, OGG, FLAC, WebM, WMATranscribed automatically
VideoMP4The audio track is extracted and transcribed
TranscriptsSRT, VTTImported as they are — no transcription, ready for analysis

Other video formats (MOV, AVI…) are not accepted — convert them to MP4 or extract the audio first.

  • Minimum duration: 10 seconds. Shorter recordings — ring-only, beep-only or empty phone-system exports — fail with Audio too short and are not billed.
  • A bitrate of at least 16 kbps is recommended for reliable results.
  • Audio is billed by its duration once transcription finishes, regardless of how many analyses you run on it afterwards.
  • Uploaded transcripts (SRT, VTT) are billed by the duration their timestamps cover, only when they are first analyzed. In a call project with automatic analysis, that is as soon as they are queued for it.

To be emailed when your uploads finish, turn on Notify job completion in your Profile.


The Status column on the Files tab shows where each file is:

StatusMeaning
UploadedReceived, transcription not started — usually because the subscription had no minutes left at upload time
TranscribingBeing transcribed. This typically takes about 1/8 of the audio’s length
TranscribedDone — the transcript can be opened, downloaded and analyzed
FailedSomething went wrong; when the reason is known it appears under the status: Audio too short or No speech detected

Call projects can also show Automatic analysis skipped — quota exceeded or Automatic analysis failed under the status.

Files with their statuses

The ⋯ menu next to each file offers:

ActionWhen it appears
TranscribeThe file is Uploaded but transcription did not start
Retry transcriptionThe file is Failed
Download → Word (DOCX) or VTTThe file is Transcribed
Edit call metadataCall projects only — see Call Metadata
DeleteAlways

When the subscription has no minutes left, Transcribe and Retry transcription read (quota exceeded) and are disabled.

Select several files with the checkboxes to Download, Transcribe or Delete them together.


Click the name of a Transcribed file to open its transcript. It is a table with one row per speaker turn:

ColumnShows
TurnThe turn’s number in the conversation
TimeWhen the turn starts and ends in the recording
SpeakerWho is speaking
TextWhat was said

Speakers are identified automatically. Labels are consistent within a file — the same speaker keeps the same label throughout — but not across files.

Transcripts are read-only in the app. To correct or annotate one, download it in Word format.

  • One file — ⋯ → Download → Word (DOCX) or VTT
  • Several files — select them, click Download and choose the Export Format. Every selected file must be Transcribed, and each one downloads as its own file.

Transcript quality drives everything built on it.

When recording

  1. Use a decent microphone and position it so every speaker is audible
  2. Minimize background noise — background music and noise noticeably reduce accuracy
  3. Make a short test recording to check levels first

Before uploading

  1. Trim long silences and irrelevant sections
  2. Prefer common formats — MP3 or M4A work best
  3. Normalize the volume if speakers are much louder or quieter than one another

After transcription

  1. Read a few transcripts before analyzing a large batch — the AI is accurate, not perfect
  2. Check the speaker split — similar voices and overlapping speech are where it goes wrong
  3. Note specialized terms and names the recognizer gets wrong; they explain odd analysis results later

Transcription is taking a long time

  • Long recordings take proportionally longer, and private transcription is slower than the default
  • Very poor audio needs more processing

A file stays Uploaded

  • Transcription did not start, usually because the subscription had no minutes left. Once minutes are available, use ⋯ → Transcribe

Poor accuracy

  • Check the audio is clear speech with little background noise or music
  • Check the recording is genuinely in speech the recognizer can hear — not a hold line or a voicemail greeting

Speakers not identified correctly

  • Very similar voices may be confused, and overlapping speech or very short turns can be attributed to the wrong speaker
  • Set Number of speakers in Project Settings when every recording has the same number of voices

A file won’t upload

Shown under FailedCauseWhat to do
Audio too shortThe recording is under 10 seconds, usually an empty or ring-only export from a phone systemCheck the exported recording; re-export the call if a real recording exists. Retrying will not help
No speech detectedThe audio contained no usable speech, so the transcript was discarded rather than inventedConfirm the recording is audible and contains speech. Re-uploading the same audio gives the same result
No reason shownA processing error, or private transcription was temporarily unavailable⋯ → Retry transcription. If it fails again, contact support