Pitch & Loudness Comparison Tool — User Guide
Dual-track analysis: compares pitch and loudness contours between two audio files frame-by-frame, designed for teacher-student comparison in vocal training and language learning.
What this does
This script implements frame-by-frame audio comparison — a specialized tool for comparing two audio recordings along pitch and loudness dimensions. The tool is designed for educational contexts where a "teacher" recording (model) is compared with a "student" recording (attempt). It performs: (1) Temporal Alignment: Aligns the two recordings by time frame based on analysis step size. (2) Loudness Analysis: Computes intensity contours, compares decibel levels frame-by-frame, calculates statistical differences. (3) Pitch Analysis: Extracts fundamental frequency contours, compares in semitone units, assesses pitch matching accuracy. (4) Comprehensive Metrics: Generates multiple difference measures including average difference, RMS difference, maximum difference, dynamic range comparison, and consistency measures.
Key Features:
- Three Analysis Modes — Loudness only, Pitch only, or Both
- Frame-by-Frame Comparison — Precise temporal alignment and comparison
- Educational Focus — Designed for teacher-student vocal training
- Multiple Difference Metrics — Average, RMS, max, min differences
- Dynamic Range Analysis — Compares overall loudness variation
- Consistency Measurement — Standard deviation comparison
- Intelligent Handling — Skips unvoiced frames in pitch analysis
- Clear Output — Formatted results in Praat Info window
Why frame-by-frame comparison? Traditional audio comparison: Overall statistics (mean, range). Frame-by-frame comparison: Temporal evolution matching. Advantages: (1) Temporal precision: Identifies where differences occur in time. (2) Pattern recognition: Detects matching of contours, not just averages. (3) Educational value: Shows students exactly where they deviate from model. (4) Detailed feedback: Multiple metrics provide comprehensive assessment. (5) Alignment insensitive: Works even if recordings aren't perfectly time-aligned (compares aligned frames). Use cases: Language learning (pronunciation coaching), singing training (pitch matching), speech therapy (prosody training), audio forensics (recording comparison), voice analysis (style imitation).
Technical Implementation: The script processes two sounds in parallel: (1) Intensity Extraction: Converts each sound to intensity contour with 75 Hz pitch floor, user-defined time step. (2) Pitch Extraction: Converts each sound to pitch contour with user-defined floor/ceiling. (3) Frame Alignment: Uses minimum number of frames between the two analyses. (4) Difference Calculation: For each aligned frame, computes absolute difference (dB for loudness, semitones for pitch). (5) Statistical Aggregation: Computes average, RMS, min, max differences across all frames. (6) Additional Metrics: Dynamic range difference (max-min), consistency difference (std deviation). Key insight: The comparison is symmetric but assumes Sound1 as reference/teacher, Sound2 as student/attempt.
Quick start
- In Praat, select EXACTLY TWO Sound objects.
- Important: First selected = Teacher/Reference, Second selected = Student/Attempt.
- Run script… →
compare_pitch_and_loudness_FIXED_STDDEV_ARGS.praat. - Choose analysis_type: Loudness_only, Pitch_only, or Both_pitch_and_loudness.
- Set time_step (default 0.01s = 100 Hz frame rate).
- For pitch analysis: Set pitch_floor_hz and pitch_ceiling_hz.
- Click OK — results appear in Info window.
- Review comparison metrics for feedback.
Quick tip: For language learning, use Both_pitch_and_loudness to analyze prosody. Set time_step = 0.01 for detailed analysis (100 frames/sec). For singing practice, adjust pitch_floor_hz and ceiling to match vocal range (e.g., 80-400 Hz for typical singing). Ensure recordings are similar duration — longer ones will be truncated to shorter duration for comparison. The script aligns by frame number, not by absolute time — slight timing differences are okay. For best results, normalize both recordings to similar loudness before comparison. Watch for "Voiced frames analyzed" count in pitch results — low counts indicate poor pitch detection (may need adjustment). Results show both absolute differences and relative patterns.
Important: EXACTLY TWO SOUNDS REQUIRED — Script will fail with 0, 1, or 3+ sounds. Order matters: First selected = Teacher, Second selected = Student — reverse order reverses interpretation. Different durations handled: Comparison uses minimum frames — longer recording truncated to match shorter. Pitch analysis requires voiced sounds — Unvoiced frames (fricatives, silence) are skipped in pitch comparison. Intensity in dB — Values are relative to Praat's reference (1e-12 W/m²), not perceptual loudness. Semitone calculation uses logarithmic formula: 12 × ln(f2/f1)/ln(2). Time step trade-off: Smaller = more frames = more detailed but slower. Pitch floor/ceiling must bracket actual pitch — incorrect settings cause pitch detection failures.
Loudness Comparison
🔊 Loudness Analysis Pipeline
Step 1: Convert each sound to intensity contour (75 Hz pitch floor)
Step 2: Extract frame-by-frame dB values
Step 3: Align frames (use minimum number)
Step 4: Compute differences for each aligned frame
Step 5: Calculate aggregate statistics
Step 6: Compute additional loudness metrics
Intensity Extraction Parameters
| Parameter | Default | Effect | Recommended Range |
|---|---|---|---|
| time_step | 0.01 s | Frame rate (100 Hz) | 0.005-0.02 s |
| Pitch floor (implied) | 75 Hz | Minimum frequency in intensity | 50-100 Hz |
| Subtract mean | yes | Removes DC component | Always yes |
Loudness Difference Metrics
FRAME-BY-FRAME DIFFERENCES:
For each frame i (1 to N):
db1[i] = intensity of teacher at frame i
db2[i] = intensity of student at frame i
if both defined:
diff[i] = |db1[i] - db2[i]|
AGGREGATE STATISTICS:
1. AVERAGE DIFFERENCE:
avg_diff = (sum of diff[i]) / N
2. RMS DIFFERENCE:
rms_diff = sqrt( (sum of diff[i]²) / N )
3. MAXIMUM DIFFERENCE:
max_diff = maximum(diff[i])
4. MINIMUM DIFFERENCE:
min_diff = minimum(diff[i])
ADDITIONAL METRICS:
5. DYNAMIC RANGE DIFFERENCE:
dr1 = max(db1) - min(db1)
dr2 = max(db2) - min(db2)
dr_diff = |dr1 - dr2|
6. CONSISTENCY DIFFERENCE:
std1 = standard_deviation(db1)
std2 = standard_deviation(db2)
consistency_diff = |std1 - std2|
Interpreting Loudness Results
| Metric | Excellent | Good | Needs Work | Interpretation |
|---|---|---|---|---|
| Average dB difference | < 2 dB | 2-5 dB | > 5 dB | Overall loudness matching |
| RMS dB difference | < 3 dB | 3-6 dB | > 6 dB | Consistent matching |
| Max dB difference | < 10 dB | 10-20 dB | > 20 dB | Worst-case deviation |
| Dynamic range diff | < 3 dB | 3-8 dB | > 8 dB | Expression matching |
| Consistency diff | < 2 dB | 2-4 dB | > 4 dB | Steadiness matching |
Educational Guidelines for Loudness
🎤 Loudness Training Scenarios
Scenario 1: Volume Matching
- Goal: Match overall loudness level
- Target: Average difference < 3 dB
- Focus: Average and RMS differences
Scenario 2: Dynamic Expression
- Goal: Match loudness contours/expression
- Target: Dynamic range difference < 5 dB
- Focus: Dynamic range and consistency differences
Scenario 3: Loudness Control
- Goal: Maintain steady volume
- Target: Consistency difference < 2 dB
- Focus: Standard deviation comparison
Common Loudness Issues and Solutions
ISSUE: Large average difference (> 5 dB)
CAUSE: Overall volume mismatch
SOLUTION: Normalize student recording to match teacher's RMS level
ISSUE: Large dynamic range difference (> 8 dB)
CAUSE: Different expressive patterns
SOLUTION: Practice crescendo/decrescendo matching
ISSUE: Large max difference (> 20 dB)
CAUSE: Sudden loud spikes or drops
SOLUTION: Identify and smooth extreme variations
ISSUE: Inconsistent matching (high RMS > low average)
CAUSE: Erratic volume changes
SOLUTION: Work on steady breath control
Pitch Comparison
🎵 Pitch Analysis Pipeline
Step 1: Convert each sound to pitch contour (user-defined range)
Step 2: Extract frame-by-frame F0 values (Hz)
Step 3: Align frames (use minimum number)
Step 4: Compute semitone differences for voiced frames only
Step 5: Calculate aggregate pitch statistics
Pitch Extraction Parameters
| Parameter | Default | Effect | Recommended Range |
|---|---|---|---|
| time_step | 0.01 s | Frame rate (100 Hz) | 0.005-0.02 s |
| pitch_floor_hz | 75 Hz | Minimum F0 to search | Male: 75-100, Female: 150-200 |
| pitch_ceiling_hz | 600 Hz | Maximum F0 to search | Male: 300-400, Female: 500-600 |
Pitch Difference Calculation
SEMITONE CALCULATION:
For two frequencies f1 and f2 (in Hz):
semitone_difference = |12 × ln(f2 / f1) / ln(2)|
Where:
• ln = natural logarithm
• 12 semitones per octave
• ln(2) ≈ 0.693147
Examples:
f1 = 220 Hz (A3), f2 = 440 Hz (A4) → 12.00 semitones (octave)
f1 = 261.6 Hz (C4), f2 = 293.7 Hz (D4) → 2.00 semitones (whole step)
f1 = 100 Hz, f2 = 105 Hz → 0.84 semitones (≈ quarter tone)
FRAME PROCESSING:
For each frame i:
f1 = teacher pitch at frame i (Hz)
f2 = student pitch at frame i (Hz)
If f1 > 0 AND f2 > 0 (both voiced):
semitone_diff[i] = |12 × ln(f2/f1)/ln(2)|
Include in statistics
Else (unvoiced in either):
Skip frame
AGGREGATE STATISTICS:
• Average semitone difference = mean(semitone_diff[i])
• Maximum semitone difference = max(semitone_diff[i])
• Voiced frames count = number of frames where both voiced
Interpreting Pitch Results
| Metric | Excellent | Good | Needs Work | Interpretation |
|---|---|---|---|---|
| Average semitone difference | < 0.5 st | 0.5-1.5 st | > 1.5 st | Overall pitch accuracy |
| Max semitone difference | < 3 st | 3-6 st | > 6 st | Worst-case pitch error |
| Voiced frames analyzed | > 80% | 50-80% | < 50% | Pitch detection reliability |
Pitch Detection and Voicing
VOICED FRAME DETECTION:
A frame is considered "voiced" if:
pitch > 0 Hz (Praat's pitch detector found F0)
Typical voiced percentages:
• Vowels: 95-100% voiced
• Nasals: 80-95% voiced
• Voiced fricatives: 50-80% voiced
• Unvoiced sounds: 0-20% voiced
• Silence: 0% voiced
IMPORTANT NOTES:
• Only voiced frames included in pitch statistics
• Unvoiced frames (fricatives, stops, silence) skipped
• Low voiced percentage may indicate:
- Incorrect pitch floor/ceiling
- Very breathy/whispered voice
- Background noise
- Technical issues with recording
Educational Guidelines for Pitch
🎶 Pitch Training Scenarios
Scenario 1: Absolute Pitch Matching
- Goal: Match exact pitch frequencies
- Target: Average difference < 0.5 semitones
- Exercises: Sustained note matching, pitch glide following
Scenario 2: Relative Pitch Contours
- Goal: Match pitch patterns/melodies
- Target: Average difference < 1.5 semitones with good contour
- Exercises: Sentence intonation, musical phrase imitation
Scenario 3: Pitch Stability
- Goal: Maintain steady pitch
- Target: Max difference < 3 semitones during sustained notes
- Exercises: Long tone practice, vibrato control
Common Pitch Issues and Solutions
ISSUE: High average semitone difference (> 2 st)
CAUSE: Systematic pitch offset (singing in wrong key)
SOLUTION: Practice with reference tones, use pitch pipe
ISSUE: Large max difference (> 6 st) but low average
CAUSE: Occasional large errors (cracking, slips)
SOLUTION: Identify problematic notes, practice transitions
ISSUE: Low voiced frames percentage (< 50%)
CAUSE: Breathy voice, incorrect pitch range settings
SOLUTION: Adjust pitch floor/ceiling, improve vocal fold closure
ISSUE: Inconsistent matching (high max, medium average)
CAUSE: Erratic pitch control
SOLUTION: Slower practice, focus on steady tones first
Result Interpretation
📊 Comprehensive Assessment Framework
Holistic View: Consider all metrics together, not individually
Context Matters: Different goals require different metric priorities
Progress Tracking: Compare metrics across practice sessions
Threshold Guidelines: Use suggested ranges as starting points
Combined Pitch and Loudness Assessment
| Performance Level | Pitch (Avg st diff) | Loudness (Avg dB diff) | Typical Profile |
|---|---|---|---|
| Expert | < 0.5 | < 2 | Near-perfect matching |
| Advanced | 0.5-1.0 | 2-3 | Good matching, minor variations |
| Intermediate | 1.0-2.0 | 3-5 | Recognizable pattern, noticeable differences |
| Beginner | 2.0-4.0 | 5-8 | Basic contour followed, significant errors |
| Novice | > 4.0 | > 8 | Little pattern matching |
Example Output Interpretation
SAMPLE OUTPUT (Both analysis):
=== LOUDNESS COMPARISON ===
Frames compared: 452
Average dB difference: 4.237 dB
RMS dB difference: 5.812 dB
Max dB difference: 18.943 dB
Min dB difference: 0.012 dB
Dynamic range difference: 7.456 dB
Consistency difference (std): 3.128 dB
=== PITCH COMPARISON ===
Voiced frames analyzed: 387 out of 452
Average semitone difference: 1.843
Max semitone difference: 8.237
INTERPRETATION:
1. Loudness: Moderate matching (4.2 dB average diff)
- Dynamic expression differs (7.5 dB range diff)
- Some consistency issues (3.1 dB std diff)
2. Pitch: Intermediate level (1.8 st average diff)
- Some large errors (8.2 st max diff)
- Good voiced detection (387/452 = 86%)
3. Overall: Student follows general patterns
- Needs work on pitch accuracy and loudness control
- Focus on reducing maximum errors
Progress Tracking Over Time
TRACKING TEMPLATE:
Session 1:
Pitch avg: 3.2 st | Loudness avg: 6.5 dB
Issues: Large pitch errors, volume mismatch
Session 2:
Pitch avg: 2.1 st | Loudness avg: 4.8 dB
Progress: Both improved, still significant differences
Session 3:
Pitch avg: 1.4 st | Loudness avg: 3.2 dB
Progress: Good improvement, approaching target ranges
Session 4:
Pitch avg: 0.9 st | Loudness avg: 2.4 dB
Goal achieved: Good matching, minor variations
KEY INSIGHTS:
• Track relative improvement, not just absolute values
• Different metrics may improve at different rates
• Plateaus are normal in skill development
• Some sessions may show regression (normal in learning)
Limitations and Considerations
Not a complete assessment: Only measures pitch and loudness — doesn't assess timbre, articulation, vowel quality, or timing precision. Frame alignment assumption: Assumes similar timing — large tempo differences will cause misalignment. Pitch detection limitations: Praat's pitch detector can fail on breathy, creaky, or noisy voices. Loudness vs perceived loudness: dB measurements don't account for equal-loudness contours (Fletcher-Munson). No normalization: Doesn't automatically adjust for different recording levels — pre-normalization recommended. Binary voiced/unvoiced: Simplified model — real voice has degrees of voicing. Semitone calculation assumes equal temperament: May not match perception for microtonal music.
Educational Applications
Language Learning and Pronunciation
🗣️ Pronunciation and Intonation Training
Goal: Improve pronunciation accuracy through prosody matching
Workflow:
- Teacher records model sentence with target intonation
- Student records imitation attempt
- Run Both_pitch_and_loudness analysis
- Review metrics together
- Identify specific problem areas
- Practice with focused exercises
Target metrics: Pitch average < 1.5 st, Loudness average < 4 dB
Singing and Vocal Training
🎤 Vocal Pitch Accuracy Training
Goal: Improve pitch accuracy for singers
Workflow:
- Select Pitch_only analysis
- Teacher sings target phrase
- Student imitates
- Analyze pitch matching
- Focus on reducing average and maximum differences
- Track progress over practice sessions
Target metrics: Professional: < 0.3 st, Student: < 1.0 st
Speech Therapy and Voice Rehabilitation
🏥 Voice Therapy Applications
Goal: Monitor voice parameter changes in therapy
Workflow:
- Record baseline (healthy model or patient's best effort)
- Record current performance
- Compare to track changes
- Use metrics as objective measures of progress
- Adjust therapy based on results
Applications: Parkinson's voice therapy, vocal fold paralysis, pitch therapy for transgender voice
Accent Reduction and Dialect Training
🌍 Prosody Pattern Acquisition
Goal: Acquire target language/dialect prosody patterns
Workflow:
- Native speaker records target prosody patterns
- Learner records imitation
- Compare using Both analysis
- Identify which aspects (pitch vs loudness) need most work
- Practice specific problematic patterns
Example: English question intonation, Mandarin tones, Japanese pitch accent
Advanced Educational Techniques
Segmented Practice Approach:
- Isolate: Practice single words or short phrases
- Analyze: Get immediate feedback on each attempt
- Correct: Adjust based on specific metrics
- Integrate: Combine into longer phrases
- Automate: Practice until metrics reach target ranges
Example progression: Single vowel → Word → Phrase → Sentence → Paragraph
Gamification Strategies:
- Score system: Convert metrics to points (lower differences = higher scores)
- Level progression: Unlock harder material as metrics improve
- Challenge modes: Try to beat previous best scores
- Multiplayer: Compare scores among students (healthy competition)
- Achievements: Unlock badges for reaching metric milestones
Troubleshooting Common Issues
Problem: Very low voiced frame count in pitch analysis
Causes: Incorrect pitch range, breathy voice, background noise, recording issues
Solutions: Adjust pitch_floor_hz and pitch_ceiling_hz, improve recording quality, use pop filter
Causes: Incorrect pitch range, breathy voice, background noise, recording issues
Solutions: Adjust pitch_floor_hz and pitch_ceiling_hz, improve recording quality, use pop filter
Problem: Extremely high dB differences (> 30 dB)
Causes: Different recording levels, clipping, normalization needed
Solutions: Normalize both recordings to similar RMS levels before comparison
Causes: Different recording levels, clipping, normalization needed
Solutions: Normalize both recordings to similar RMS levels before comparison
Problem: Inconsistent results across attempts
Causes: Variable performance, different recording conditions, analysis parameter changes
Solutions:
Causes: Variable performance, different recording conditions, analysis parameter changes
Solutions:
Problem: Analysis takes very long
Causes: Long recordings, very small time_step
Solutions: Use shorter segments for practice, increase time_step (e.g., 0.02s), focus on key phrases
Causes: Long recordings, very small time_step
Solutions: Use shorter segments for practice, increase time_step (e.g., 0.02s), focus on key phrases
Integration with Other Tools
| Tool | Integration | Enhanced Capability |
|---|---|---|
| Praat TextGrid | Segment recordings by phoneme/word | Compare specific speech segments |
| Spreadsheet software | Export metrics for tracking | Long-term progress visualization |
| Audio editor | Pre-process recordings | Normalize, trim, filter before analysis |
| Recording app | Standardize recording setup | Consistent input quality |
| Learning management system | Embed in online courses | Distance learning applications |