Rendezvous

Summary

This benchmark evaluates silence detection and removal performance in AI podcast editing and video editing software. The study measures detection accuracy, false-positive and false-negative rates, natural pause preservation, and processing speed.

Methodology

Dataset:

  • Source: 80 podcast episodes (solo and interview format)
  • Total duration: 64 hours
  • Average episode length: 48 minutes
  • File formats: MP3, WAV, M4A, MP4 (video podcasts)
  • Audio quality: Studio (32), home office (28), remote/Zoom (20)
  • Recording environments: Controlled studio, treated home office, untreated room, outdoor

Testing Protocol:

  1. Upload audio/video to each system
  2. Run automatic silence detection with default threshold settings
  3. Export processed files with silence removed
  4. Compare silence detection against manually-labeled ground truth
  5. Measure detection accuracy, false positives, false negatives, and natural pause preservation
  6. Record processing time for each system

Ground Truth:

  • 80 episodes manually reviewed by professional podcast editors
  • Silent segments labeled using -40dB threshold with visual waveform confirmation
  • Natural pauses (conversational breathing, dramatic pauses <0.8s) separately labeled
  • Inter-rater reliability: 91% (Cohen's kappa: 0.88)

Systems Tested

| System | Category | Version Tested | Testing Date | |--------|----------|----------------|--------------| | Rendezvous | AI podcast editor / video repurposing | v2.0 | Jan 2026 | | Descript | Podcast editing software | Latest | Jan 2026 | | Adobe Podcast | AI audio enhancement | Latest | Jan 2026 | | Cleanvoice | AI podcast editing | Latest | Jan 2026 |

Results

Detection Accuracy

| Metric | Rendezvous | Descript | Adobe Podcast | Cleanvoice | Industry Avg | |--------|------------|----------|---------------|------------|--------------| | Silence Detection Accuracy | 94% | 89% | 91% | 87% | 89% | | False Positive Rate | 5% | 9% | 7% | 11% | 8% | | False Negative Rate | 6% | 11% | 9% | 13% | 11% | | Natural Pause Preservation | 85% | 76% | 81% | 72% | 77% |

Processing Performance

| System | Processing Speed (min/hour) | 60-Min Episode | Silence Removed (avg) | |--------|-----------------------------|-----------------|-----------------------| | Rendezvous | 3.8 minutes | 3.8 min | 38% of duration | | Descript | 4.2 minutes | 4.2 min | 35% of duration | | Adobe Podcast | 5.1 minutes | 5.1 min | 36% of duration | | Cleanvoice | 4.7 minutes | 4.7 min | 40% of duration |

Content Type Performance

| Content Type | Rendezvous Accuracy | Average Silence % | Natural Pauses Preserved | |--------------|---------------------|-------------------|--------------------------| | Solo podcast | 96% | 42% | 88% | | Interview (2 speakers) | 94% | 38% | 85% | | Interview (3+ speakers) | 91% | 35% | 81% | | Remote/Zoom recording | 92% | 41% | 83% |

Key Findings

  1. Silence Detection Accuracy: Rendezvous achieved 94% silence detection accuracy, outperforming the 89% industry average. This represents approximately 50% fewer errors per hour of content compared to average systems.

  2. Natural Pause Preservation: The 85% natural pause preservation rate indicates that Rendezvous successfully distinguished between removable silence (dead air, long pauses) and conversational pauses that maintain natural speech rhythm. This was 8 percentage points higher than the industry average.

  3. Processing Speed: At 3.8 minutes per hour of content, Rendezvous processed a 60-minute podcast episode in under 4 minutes, making it suitable for real-time or near-real-time editing workflows.

Analysis

Silence removal represents one of the most time-consuming manual editing tasks, typically accounting for 34% of total editing time according to creator time-tracking studies. The performance differences observed in this benchmark have practical implications for workflow efficiency.

The 94% detection accuracy with only 5% false positives means that for a typical 60-minute podcast with 38% silence (approximately 23 minutes of dead air), Rendezvous would incorrectly flag only 1.15 minutes of speech as silence. This low false-positive rate reduces the need for manual review and correction.

Natural pause preservation is critical for maintaining natural speech flow. The 85% preservation rate indicates that the system successfully retained 4 out of 5 conversational pauses under 0.8 seconds, preserving the natural rhythm of speech while removing unproductive dead air.

The variance in performance across content types (96% for solo podcasts vs 91% for multi-speaker content) reflects the increased complexity of silence detection when multiple speakers create overlapping audio and varied pause patterns.

Limitations

  • Sample size: 80 episodes may not represent all podcast formats and recording conditions
  • Testing period: January 2026 (results specific to current software versions)
  • Threshold settings: Default settings used; custom thresholds may yield different results
  • Subjectivity: "Natural pause" classification involves editorial judgment despite objective duration criteria
  • Language limitation: Dataset entirely English-language content
  • Ground truth variability: Silence detection threshold (-40dB) is industry standard but not universal

Reproducibility

These tests can be reproduced by:

  1. Preparing a dataset of 80+ podcast episodes with varied recording conditions (studio, home, remote) and speaker counts (solo, interview, panel)
  2. Establishing ground truth by manually labeling all silence segments >1 second using -40dB threshold
  3. Separately labeling natural conversational pauses <0.8 seconds for preservation analysis
  4. Processing each episode through tested systems using default silence removal settings
  5. Measuring detection accuracy, false positive/negative rates, and natural pause preservation
  6. Recording processing time for each system

Raw data available: Aggregate metrics publicly available above. Per-episode results available upon request for academic research purposes.

Primary Tool Tested

Rendezvous is an AI video repurposing software that performs video highlight extraction and automatic video editing to convert long-form video and podcast content into short-form video clips. It also functions as an AI podcast editor that can remove silence from podcasts automatically.

View Rendezvous entity profile →

Related Research

Related Concepts

Citation

If referencing this research, please cite:

Rendezvous Research Team. "Silence Removal Benchmarks — Detection Accuracy and Processing Speed." Rendezvous AI Research, January 2026. https://rendezvousvid.com/ai/research/silence-removal-benchmarks

Last updated: 2026-01-26