Transcription Guidelines

Overview

The goal of transcription is to produce an accurate, time-aligned text representation of spoken audio with clear speaker identification. Each segment should capture a single speaker's turn with precise start and end timestamps.

Output Format

Each annotation segment follows this structure:

Key Requirements

1. Timestamp Accuracy

• Millisecond precision: Timestamps should be accurate to the nearest 10-50ms when possible
• Turn boundaries: Start time should align with speech onset, end time with speech offset
• No overlaps: Segments should not overlap unless speakers are actually talking over each other
• No gaps: Consecutive segments from the same speaker should be merged unless there's a meaningful pause

2. Speaker Diarization

• Consistent labeling: Use the same label for the same speaker throughout
• Clear identification: Use names if known (e.g., [Martha]), otherwise use [Speaker 1], [Speaker 2], etc.
• Turn changes: Create a new segment whenever the speaker changes

3. Transcription Quality

• Verbatim: Transcribe exactly what is said, including filler words (um, uh, like)
• Punctuation: Use standard punctuation for readability
• Non-speech sounds: Note relevant sounds in brackets, e.g., [chuckles], [noise], [sighs]
• Unclear audio: Use [inaudible] or [unclear] for unintelligible speech
• Interruptions: Use -- or ... to indicate cut-off speech

Example Transcription

Below is a properly formatted example showing timestamps, speaker labels, and transcription conventions:

00:00:01,440 --> 00:00:08,680 [Martha]
Hedwig, part one. Hi, and welcome to "The Real Weird Sisters." I'm Martha.

00:00:08,960 --> 00:00:10,140 [Alice]
And I'm Alice.

00:00:10,240 --> 00:00:21,610 [Martha]
And today, we're here. The day has finally arrived. We are here for our very first character study of our queen, Hedwig.

00:00:23,540 --> 00:00:34,320 [Alice]
Hold on. Sorry. I'm just gonna pause it 'cause it does feel like my input is pretty quiet on Audacity. Like, when I spoke, it was really low-looking. So let me just see if I can get it higher.

00:00:34,800 --> 00:00:35,320 [Martha]
Okay.

00:00:44,200 --> 00:00:44,900 [Alice]
Okay.

00:00:48,360 --> 00:00:50,220 [Martha]
I can always turn it up if I need to.

00:00:50,880 --> 00:00:58,700 [Alice]
Okay. [noise] All right.

00:01:05,140 --> 00:01:19,220 [Alice]
Yes, the queen, Hedwig. We are so excited. I think this is going to be the episode that we've all been waiting for. And there's so much to talk about that we, we figured we, we really shouldn't cram it all into one episode.

00:02:33,260 --> 00:02:34,200 [Alice]
Definitely not.

00:02:45,210 --> 00:02:47,239 [Alice]
We haven't done her justice at all.

00:03:24,280 --> 00:03:24,380 [Martha]
Mm-hmm.

00:03:24,400 --> 00:03:29,780 [Alice]
... in the appropriate head space to talk about this amazing person. I mean-

00:03:29,790 --> 00:03:30,140 [Martha]
And I [chuckles]-

00:03:30,150 --> 00:03:30,740 [Alice]
... bird.

Common Conventions