Garchen Rinpoche STT Benchmark Dataset Report

Generated on: 2025-06-30 12:03:35

Total catalog size: 38 original audio files Total segments: 13786 audio segments Valid segments (duration filtered): 13786 audio segments Benchmark size: 978 audio segments

Sampling Strategy

Stratified sampling was used to ensure representation across:

Age groups
Duration categories
Content types

Target segments per stratum: 130 Duration filter range: 0.5-30.0 seconds

Distribution of Strata in Original Dataset

Age Groups

70-80: 33 (86.8%)
80-90: 5 (13.2%)

Duration Categories

long: 20 (52.6%)
medium: 14 (36.8%)
short: 4 (10.5%)

Content Types

Teaching: 25 (65.8%)
Practice: 5 (13.2%)
Q&A: 4 (10.5%)
Prayer: 4 (10.5%)

Distribution in Valid Segments

Total unique strata: 8

80-90__long__Practice: 8757 segments (63.5%)
70-80__long__Teaching: 2279 segments (16.5%)
70-80__medium__Teaching: 1085 segments (7.9%)
70-80__long__Q&A: 704 segments (5.1%)
70-80__long__Prayer: 479 segments (3.5%)
70-80__short__Teaching: 213 segments (1.5%)
70-80__medium__Prayer: 201 segments (1.5%)
70-80__medium__Q&A: 68 segments (0.5%)

Distribution in Benchmark Selection

Age Groups

70-80: 848 (86.7%)
80-90: 130 (13.3%)

Duration Categories

long: 520 (53.2%)
medium: 328 (33.5%)
short: 130 (13.3%)

Content Types

Teaching: 390 (39.9%)
Prayer: 260 (26.6%)
Q&A: 198 (20.2%)
Practice: 130 (13.3%)

Segment Duration Statistics

Average duration: 4.20 seconds
Minimum duration: 0.51 seconds
Maximum duration: 29.90 seconds