cvanal

Convolution analysis utility

The cvanal utility performs FFT analysis on impulse response files for use with Csound’s convolution opcodes. It converts time-domain impulse responses into frequency-domain representations optimized for fast convolution.

Syntax

cvanal [options] input_soundfile output_analysis_file

Arguments

• input_soundfile - Input impulse response audio file
• output_analysis_file - Output convolution analysis file

Options

File and time selection

Option	Description
-s <rate>	Override input sample rate
-c <channel>	Channel to analyze (1-4, default: all channels)
-b <time>	Begin time in seconds
-d <duration>	Duration to analyze in seconds

File format

Option	Description
-X	Use new text-based format (machine-independent)

Examples

Basic impulse response analysis

cvanal impulse.wav impulse.cv

Analyze specific channel

cvanal -c 1 stereo_ir.wav left_ir.cv

Analyze time segment

cvanal -b 0.5 -d 2.0 long_ir.wav short_ir.cv

Use portable text format

cvanal -X impulse.wav impulse.cv

Multi-channel room response

cvanal room_response.wav room.cv

Output file format

The analysis file contains:

• Header: Sample rate, channels, impulse response length, FFT size
• FFT data: Frequency-domain representation of the impulse response
• Format flag: Indicates rectangular (time-domain derived) format

FFT size and zero-padding

The utility automatically:

1. Determines impulse response length H from input
2. Calculates FFT size as next power of 2 ≥ 2H - 1
3. Zero-pads impulse response to FFT size
4. Performs FFT on zero-padded signal

This ensures proper convolution without time-domain aliasing.

Impulse response considerations

Recording impulse responses

Good impulse responses should:

• Start at time zero with no pre-delay
• Decay to silence within the recording
• Have minimal background noise
• Be normalized to prevent clipping

Types of impulse responses

Room acoustics:

• Concert halls
• Studios
• Churches
• Special acoustic spaces

Device responses:

• Speaker cabinets
• Microphone characteristics
• Analog equipment
• Audio effects

Synthetic responses:

• Custom filter designs
• EQ curves
• Artificial reverbs

Channel handling

Mono impulse responses

Apply same response to all channels:

cvanal -c 1 mono_ir.wav ir.cv

Stereo impulse responses

Convolve with stereo files:

cvanal stereo_ir.wav stereo.cv

Multi-channel responses

For surround sound or specialized routing:

cvanal -c 2 multi_ir.wav channel2.cv

File formats

Binary format (default)

• Machine-dependent byte order
• Smaller file size
• Faster to read
• May not be portable across systems

Text format (-X flag)

• Machine-independent
• Portable across platforms
• Human-readable (for debugging)
• Larger file size
• Slower to read

Using convolution files in Csound

Convolution analysis files are used with:

• convolve - Time-domain convolution
• pconvolve - Partitioned convolution for efficiency

Example Csound code:

instr 1
  asig soundin "input.wav"
  aconv pconvolve asig, "room.cv"
  out aconv
endin

Performance considerations

FFT size impact

• Larger impulse responses require larger FFT sizes
• FFT size affects convolution efficiency
• Very long responses (>1 second) may require partitioned convolution

Memory usage

Memory required (bytes) ≈ FFT_size × channels × 8

For a 2-second impulse at 44.1kHz:

• Impulse length: 88,200 samples
• FFT size: 131,072 (next power of 2 ≥ 176,399)
• Memory: ~1 MB per channel

Troubleshooting

Analysis fails

• Check input file is valid audio
• Verify sufficient disk space
• Check file permissions

Output sounds wrong

• Ensure impulse response starts at time zero
• Check for DC offset in impulse (can cause artifacts)
• Verify impulse doesn’t clip
• Try normalizing impulse response first

File too large

• Trim silence from end of impulse
• Use -d to limit duration
• Consider shorter impulse responses

Technical details

FFT convolution

The frequency-domain convolution theorem states:

convolution(x, h) = IFFT(FFT(x) × FFT(h))

This is more efficient than time-domain convolution for long impulses.

Zero-padding requirement

To avoid circular convolution artifacts:

• Minimum FFT size = input_length + impulse_length - 1
• Utility rounds up to next power of 2

Related utilities

• pvanal - Phase vocoder analysis
• sndinfo - Display sound file information