Outline | SoferAi | Documentation

Outline of topics discussed by timestamp, generated end-to-end from a transcription ID.

This endpoint will:

Fetch the transcript and word-level timestamps for the given transcription
Generate chapter topics (title + starting_phrase) using an LLM from the transcript text
Align each topic’s starting phrase to timestamps

Outline of topics discussed by timestamp, generated end-to-end from a transcription ID. This endpoint will: 1) Fetch the transcript and word-level timestamps for the given transcription 2) Generate chapter topics (title + starting_phrase) using an LLM from the transcript text 3) Align each topic's starting phrase to timestamps

Authentication

AuthorizationBearer

Bearer authentication of the form Bearer <token>, where token is your auth token.

Request

This endpoint expects an object.

transcription_idUUIDRequired

ID of the transcription to process end-to-end

monotonebooleanOptional

If true, each topic is searched after the previous topic’s start (with a small backoff)

conclusion_biasbooleanOptional

If true and a title includes the word "conclusion", search in the last third of the audio

Response

This endpoint returns a list of objects.

titlestring

Title of the topic

starting_phrasestring

Starting phrase that was searched for

matched_textstring

The transcript text segment best matching the starting phrase

score_normdouble

Normalized alignment score in [0,1]

confidenceenum

Confidence bucket derived from the normalized score

Allowed values:

start_time_secdouble or null

Start time in seconds for the best-aligned phrase (if found)

token_spanlist of integers or null

Optional span of token indices [start, end) for the best match

matched_contentinteger or null

Number of matched content tokens (excludes stopwords/fillers)

content_coveragedouble or null

Fraction of content tokens in the phrase that were matched

longest_content_run_softinteger or null

Longest soft contiguous run of content-token matches

longest_content_run_strictinteger or null

Longest strict contiguous run of content-token matches

1	from soferai import SoferAI
2	from uuid import UUID
3
4	client = SoferAI(
5	api_key="YOUR_TOKEN_HERE"
6	)
7
8	client.timestamps.outline(
9	transcription_id=UUID("d5e9c84f-c2b2-4bf4-b4b0-7ffd7a9ffc32")
10	)

1	[
2	{
3	"title": "Introduction to Machine Learning",
4	"starting_phrase": "In this session, we will explore the basics of machine learning",
5	"matched_text": "In this session, we will explore the basics of machine learning including supervised and unsupervised methods.",
6	"score_norm": 0.95,
7	"confidence": "HIGH",
8	"start_time_sec": 0,
9	"token_span": [
10	0,
11	15
12	],
13	"matched_content": 15,
14	"content_coverage": 0.85,
15	"longest_content_run_soft": 12,
16	"longest_content_run_strict": 10
17	},
18	{
19	"title": "Deep Learning Techniques",
20	"starting_phrase": "Next, we dive into deep learning and neural networks",
21	"matched_text": "Next, we dive into deep learning and neural networks, discussing architectures like CNNs and RNNs.",
22	"score_norm": 0.92,
23	"confidence": "HIGH",
24	"start_time_sec": 600.5,
25	"token_span": [
26	120,
27	135
28	],
29	"matched_content": 15,
30	"content_coverage": 0.8,
31	"longest_content_run_soft": 14,
32	"longest_content_run_strict": 13
33	}
34	]

Authentication

Request

Response

Errors