Wavecap Transcripts

WaveCap Transcripts Analysis Skill

Use this skill to inspect, analyze, and audit transcription quality in WaveCap.

Get Transcriptions

List recent transcriptions for a stream

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=20" | \
  jq '.transcriptions[] | {id, timestamp, text, confidence, reviewStatus}'

Get all transcriptions across streams

curl -s http://localhost:8000/api/transcriptions/export | jq 'length'

Get transcription with full details

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=100" | \
  jq '.transcriptions[] | select(.id == "TRANSCRIPTION_ID")'

Quality Analysis

Find low-confidence transcriptions

# Very low confidence (< 0.5) - likely errors
curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.confidence != null and .confidence < 0.5)] | sort_by(.confidence) | .[:20] | .[] | {id, streamId, confidence, text}'

# Medium-low confidence (0.5 - 0.7) - may need review
curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.confidence != null and .confidence >= 0.5 and .confidence < 0.7)] | length'

Find very short transcriptions (might be noise)

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select((.text | length) < 10)] | .[:20] | .[] | {id, text, duration, confidence}'

Find transcriptions with common error patterns

# Find transcriptions with repeated words (stuttering artifact)
curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.text | test("\\b(\\w+)\\s+\\1\\b"))] | .[:10] | .[] | {id, text}'

# Find transcriptions with "[inaudible]" or similar markers
curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.text | test("inaudible|unclear|unintelligible"; "i"))] | length'

Analyze confidence distribution

curl -s http://localhost:8000/api/transcriptions/export | \
  jq 'group_by(.confidence | . * 10 | floor / 10) | map({confidence_bucket: .[0].confidence | . * 10 | floor / 10, count: length}) | sort_by(.confidence_bucket)'

Compare Original vs Corrected

List corrected transcriptions

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.reviewStatus == "corrected")] | .[] | {id, original: .text, corrected: .correctedText, reviewer: .reviewedBy}'

Show diff between original and corrected

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.reviewStatus == "corrected")] | .[:5] | .[] | "--- Original ---\n" + .text + "\n--- Corrected ---\n" + .correctedText + "\n"'

Calculate correction rate

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '{
    total: length,
    corrected: [.[] | select(.reviewStatus == "corrected")] | length,
    verified: [.[] | select(.reviewStatus == "verified")] | length,
    pending: [.[] | select(.reviewStatus == "pending" or .reviewStatus == null)] | length
  } | . + {correction_rate: (.corrected / .total * 100 | . * 100 | round / 100)}'

Review Status Audit

Count by review status

curl -s http://localhost:8000/api/transcriptions/export | \
  jq 'group_by(.reviewStatus // "pending") | map({status: .[0].reviewStatus // "pending", count: length})'

List transcriptions needing review (low confidence, unreviewed)

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select((.reviewStatus == "pending" or .reviewStatus == null) and .confidence != null and .confidence < 0.7)] | sort_by(.confidence) | .[:20] | .[] | {id, streamId, confidence, text}'

Find recently reviewed transcriptions

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.reviewedAt != null)] | sort_by(.reviewedAt) | reverse | .[:10] | .[] | {id, reviewStatus, reviewedAt, reviewedBy}'

Stream-by-Stream Analysis

Transcription count per stream

curl -s http://localhost:8000/api/transcriptions/export | \
  jq 'group_by(.streamId) | map({stream: .[0].streamId, count: length}) | sort_by(.count) | reverse'

Average confidence per stream

curl -s http://localhost:8000/api/transcriptions/export | \
  jq 'group_by(.streamId) | map({
    stream: .[0].streamId,
    count: length,
    avg_confidence: ([.[] | .confidence // 0] | add / length | . * 100 | round / 100)
  }) | sort_by(.avg_confidence)'

Find streams with quality issues

# Streams with many low-confidence transcriptions
curl -s http://localhost:8000/api/transcriptions/export | \
  jq 'group_by(.streamId) | map({
    stream: .[0].streamId,
    total: length,
    low_confidence: [.[] | select(.confidence != null and .confidence < 0.6)] | length
  }) | map(. + {pct_low: (.low_confidence / .total * 100 | round)}) | sort_by(.pct_low) | reverse | .[:10]'

Segment Analysis

Inspect transcription segments (word-level timing)

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=5" | \
  jq '.transcriptions[0].segments'

Find segments with low confidence (within a transcription)

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=10" | \
  jq '.transcriptions[] | {id, low_conf_segments: [.segments[]? | select(.no_speech_prob > 0.5 or .avg_logprob < -1)] | length}'

Time-Based Analysis

Transcriptions in the last hour

SINCE=$(date -u -v-1H +"%Y-%m-%dT%H:%M:%SZ" 2>/dev/null || date -u -d "1 hour ago" +"%Y-%m-%dT%H:%M:%SZ")
curl -s http://localhost:8000/api/transcriptions/export | \
  jq --arg since "$SINCE" '[.[] | select(.timestamp >= $since)] | length'

Activity over time

curl -s http://localhost:8000/api/transcriptions/export | \
  jq 'group_by(.timestamp | split("T")[0]) | map({date: .[0].timestamp | split("T")[0], count: length}) | sort_by(.date) | .[-14:]'

Text Search

Search transcriptions for keywords

curl -s "http://localhost:8000/api/transcriptions/search?q=fire&limit=20" | \
  jq '.results[] | {id, streamId, timestamp, text, matchCount}'

Find transcriptions mentioning addresses

curl -s http://localhost:8000/api/transcriptions/export | \
  jq '[.[] | select(.text | test("\\d+\\s+\\w+\\s+(street|st|avenue|ave|road|rd|drive|dr|lane|ln)"; "i"))] | .[:10] | .[] | {id, text}'

Tips

• Use .confidence to prioritize which transcriptions need review
• The reviewStatus field can be: pending, corrected, or verified
• correctedText contains human-corrected version when reviewStatus == "corrected"
• Segments provide word-level timing and confidence via no_speech_prob and avg_logprob
• Very short transcriptions with low confidence are often noise, not real speech
• Use the search API for keyword queries rather than filtering exports client-side