Function detectTranscriptFormat

lib/transcript-format-detector.ts:23–59

(
  transcript: TranscriptSegment[]
)

 * For edge cases (15-80% punctuation ratio), uses average text length as tiebreaker.
 */
export function detectTranscriptFormat(
  transcript: TranscriptSegment[]
): 'old' | 'new' {
  if (!transcript || transcript.length === 0) {
    return 'new'; // Default to new format for empty transcripts
  }

  // Sample first 100 segments for analysis (or all if fewer)
  const sampleSize = Math.min(100, transcript.length);
  const sample = transcript.slice(0, sampleSize);

  // Calculate average text length
  const totalTextLength = sample.reduce((sum, seg) => sum + (seg.text?.length || 0), 0);
  const avgTextLength = totalTextLength / sampleSize;

  // Calculate percentage of segments ending with sentence punctuation
  const sentenceEndingRegex = /[.!?\u3002\uff01\uff1f\u203c\u2047\u2048]\s*$/;
  const sentenceEndingCount = sample.filter(seg =>
    sentenceEndingRegex.test(seg.text?.trim() || '')
  ).length;
  const sentenceEndingRatio = sentenceEndingCount / sampleSize;

  // Prioritize punctuation ratio as the primary indicator
  // OLD format: Most segments break mid-sentence (low punctuation ratio)
  if (sentenceEndingRatio < 0.15) {
    return 'old';
  }

  // NEW format: Most segments end at sentence boundaries (high punctuation ratio)
  if (sentenceEndingRatio > 0.80) {
    return 'new';
  }

  // Medium punctuation ratio (15-80%): Use text length as tiebreaker
  // Longer segments more likely to be merged sentences
  return avgTextLength > 40 ? 'new' : 'old';
}

/**
 * Ensures transcript is in the new merged sentence format.