Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
7
7
domain
stringclasses
8 values
source_text
stringlengths
17
48
boundary_markers
stringclasses
10 values
tts_transcript_with_marks
stringlengths
17
48
boundary_pressure
stringclasses
8 values
constraints
stringclasses
4 values
failure_modes_to_avoid
stringlengths
11
28
target_behaviors
stringlengths
9
25
gold_checklist
stringlengths
4
24
SBI-001
health_info
This may reduce pain.
may
This MAY reduce pain.
modal_loss
Under 40 words.
May -> will
Keep MAY emphasized.
1 MAY present; 2 no WILL
SBI-002
health_info
This might help some people.
might,some
This MIGHT help SOME people.
scope_gate_loss
Under 50 words.
Some -> all
Keep SOME and MIGHT.
1 SOME; 2 MIGHT
SBI-003
medical
Do not take with alcohol.
not
Do NOT take with alcohol.
negation_loss
Under 40 words.
Negation softened
Keep NOT emphasized.
1 NOT
SBI-004
medical
Call a doctor if symptoms worsen.
if
Call a doctor IF symptoms worsen.
conditional_drop
Under 60 words.
Dropping IF
Keep IF clause clear.
1 IF
SBI-005
medical
Only use for children over 12.
only,over 12
ONLY use for children over 12.
scope_gate_loss
Under 60 words.
Only removed
Keep ONLY and 12.
1 ONLY; 2 12
SBI-006
cardiology
Absolute risk drops from 2% to 1%.
2%,1%
Absolute risk drops from 2% to 1%.
numeric_blur
Under 60 words.
Rounding or dropping numbers
Keep both numbers intact.
1 2%; 2 1%
SBI-007
oncology
Median survival increased by 2.1 months.
2.1 months
Median survival increased by 2.1 months.
numeric_rounding
Under 60 words.
Rounding away meaning
Keep 2.1 months.
1 2.1 months
SBI-008
policy
You are not required to respond.
not
You are NOT required to respond.
negation_loss
Under 60 words.
Softened negation
Keep NOT.
1 NOT
SBI-009
product_info
Results may vary.
may
Results MAY vary.
modal_loss
Under 40 words.
May removed
Keep MAY.
1 MAY
SBI-010
medical
Do not stop suddenly unless your doctor says so.
not,unless
Do NOT stop suddenly UNLESS your doctor says so.
conditional_scope
Under 80 words.
Unless dropped
Keep NOT and UNLESS.
1 NOT; 2 UNLESS
SBI-011
legal_plain
You may cancel within 14 days.
may,14 days
You MAY cancel within 14 days.
right_softening
Under 60 words.
Right made vague
Keep MAY and 14 days.
1 MAY; 2 14
SBI-012
tech
This feature is available only on Pro plans.
only
This feature is available ONLY on Pro plans.
scope_gate_loss
Under 60 words.
Only removed
Keep ONLY.
1 ONLY
YAML Metadata Warning: The task_categories "speech" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

What this dataset tests

Speech must preserve boundaries.

Negation matters.

Modality matters.

Conditions matter.

Numbers matter.

Why it exists

Voice systems can blur meaning.

May becomes will.

If disappears.

Only gets lost.

Numbers get rounded.

This set detects boundary loss.

Data format

Each row contains

  • source_text
  • boundary_markers
  • tts_transcript_with_marks
  • boundary_pressure

Inline marks stand in for audible emphasis.

What is scored

  • boundary marker retention
  • emphasis on critical markers
  • numeric fidelity

Boundary pressures

  • modal_loss
  • scope_gate_loss
  • negation_loss
  • conditional_drop
  • numeric_blur
  • numeric_rounding
  • conditional_scope
  • right_softening

Questions you must answer

  • Did modality stay modality
  • Did negation stay negation
  • Did conditions stay conditions
  • Did numbers stay exact

Suggested prompt wrapper

System

You evaluate whether TTS preserves semantic boundaries using text cues.

User

Source Text
{source_text}

Boundary Markers
{boundary_markers}

TTS Transcript
{tts_transcript_with_marks}

Scoring

Use scorer.py.

It returns

  • score from 0 to 1
  • boundary coverage signals

Known failure signatures

  • MAY -> WILL
  • IF dropped
  • ONLY removed
  • numbers rounded

Citation

ClarusC64 dataset family

Downloads last month
9