Skip to content
Annotask

04Service

Audio annotation

Transcripts and labeled regions for speech and sound models.

We transcribe and segment speech and sound with timestamps, following your style guide for fillers, numbers and overlapping speech. Accents and code-switching are flagged rather than guessed.

Request a QuoteSee it labeled
Annotation types
4 types
Delivered as
JSON, SRT or CSV with timestamps
Quality levels
Attempt → Review → Final Review
Starts with
A paid pilot of 500–1,000 items

01Showcase

See it labeled.

Speaker regions and a time-aligned transcript.

call_0381.wav · 2 speakers0.0s / 12.0s
Speaker A
Speaker B
noise
Speaker A
Plays the annotation timeline · no sound
  1. 0.4–3.9AHi, I'd like to change my delivery address.
  2. 4.1–6.0BSure, can I have your order number?
  3. 6.4–8.3—[hold music]
  4. 8.6–11.6AIt's four, two, seven, one, nine.

02Annotation types

What we label.

Each type is defined in your guidelines before the pilot, with examples of the edge cases.

Verbatim or clean-read transcripts, time-aligned to the audio.

Who spoke when, with a consistent label for each speaker.

Start and end times for non-speech sounds, such as alarms, music or background noise.

Language, dialect or accent labels per segment, including code-switched speech.

Studio microphone

03What you receive

Labels your pipeline can load as-is.

Delivered as JSON, SRT or CSV with timestamps. Tell us your schema and we'll match it.

  • Labels in the format your pipeline already reads
  • A quality report per batch: accuracy on hidden test items against your target
  • Every task through Attempt, Review and Final Review
  • A 7-day window to accept each batch or raise a dispute
JSONExample
{
  "file": "call_0381.wav",
  "segments": [
    { "start": 0.42, "end": 3.87, "speaker": "A", "text": "Hi, I'd like to change my delivery address." },
    { "start": 4.10, "end": 6.02, "speaker": "B", "text": "Sure, can I have your order number?" }
  ]
}

04Use cases

Where teams use it.

  1. 01

    Speech recognition

  2. 02

    Call analytics

  3. 03

    Voice assistants

05Quality

Every task passes three levels.

Only tasks approved at Final Review are paid to the annotator and delivered to you.

  1. Level 1Annotator

    Attempt

    Labels the item to the project guidelines and label schema.

  2. Level 2Reviewer

    Review

    Checks the work, then approves it, corrects it, or sends it back with a reason.

  3. Level 3Final reviewer

    Final Review

    Signs the task off. Only finally approved tasks are paid and delivered.

06Other services

Need more than one data type?

Industrial robot arm with a gripper

Image annotation

Labels for computer vision models, from simple tags to pixel-level masks.

Small wheeled robot built on an Arduino board

Video annotation

Frame-accurate labels that follow objects through a clip.

Hands typing on a laptop keyboard

Text annotation

Structured labels for NLP, search and chatbot training.

Self-driving electric shuttle with a roof-mounted sensor pod

3D and LiDAR annotation

Labels on point clouds and sensor-fusion scenes.

Monitor showing line charts and data tables

Data collection and validation

New data gathered to spec, and existing labels checked and corrected.

Start a audio annotation pilot. Judge us on your own data.

We label 500–1,000 of your items through all three levels. You approve the results before production begins.

Request a Quote