hardware.md - tphakala/birdnet-go GitHub Wiki

BirdNET-Go Hardware Recommendations

This document outlines the recommended hardware configurations for running BirdNET-Go effectively, ensuring optimal performance for both detection capabilities and user interface responsiveness.

Compute Hardware

Recommended Systems

  • Primary Recommendation: Raspberry Pi 5 (2GB RAM model is sufficient)

    • Rationale: While BirdNET-Go's core real-time detection can run adequately on older models, the Pi 5 offers significantly improved performance particularly for the web interface, resulting in much more responsive spectrogram rendering and overall experience.
    • Processing Speed: The Pi 5 processes audio chunks significantly faster than older models, allowing for more reliable Deep Detection when using high overlap settings.

  • Alternative Systems:

    • Raspberry Pi 4B (4GB) - Good performance, especially when overclocked to 1.8GHz
    • Intel NUC - Excellent performance, higher power consumption

  • Minimum Viable System: Raspberry Pi 3B+ (512MB) or Raspberry Pi Zero 2 (512MB)

    • Note: While the Pi 3B+ can handle core detection (processing 3-second segments in approximately 500ms), expect slower web interface responsiveness and possible performance issues with Deep Detection and multiple simultaneous RTSP streams.

RAM Requirements

  • Baseline Usage: ~250MB for core BirdNET-Go processes
  • Recommended:
    • 2GB RAM for standard installations (single audio source)
    • Each RTSP stream requires an additional FFmpeg process, which can consume substantial memory, if you have high number of RTSP source streams consider system with 4GB or more RAM

CPU Considerations

  • Threads: BirdNET-Go can utilize multiple CPU cores. By default, it automatically optimizes for P-cores on hybrid architectures.
  • Performance Impact: CPU power directly affects:
    • Maximum achievable overlap values (Deep Detection capability)
    • Spectrogram rendering speed
    • Multiple stream processing capacity
    • Audio export encoding speed

Operating System

  • Recommended: Raspberry Pi OS Lite (64-bit) based on Debian Bookworm (Debian 12)

    • Rationale:
      • 64-bit OS is required (not optional) because TensorFlow Lite requires a 64-bit system
      • "Lite" version avoids unnecessary desktop components, saving resources
      • Bookworm provides current package support for dependencies

  • Alternative OS Options:

    • Ubuntu Server 22.04 LTS or newer (64-bit)
    • Debian 12 or newer (64-bit)
    • For desktop systems: Windows 10/11, macOS

Audio Input Hardware

Sound Cards

  • Recommended: USB audio interfaces with the following characteristics:
    • 48kHz sample rate support
    • Low self-noise
    • Proper line/mic input with appropriate gain control
    • Models to consider:
      • CM108 based USB sound card
      • Creative Sound Blaster Play! 3
      • U-Green USB sound card (multiple models)

Microphones

Important Note: BirdNET AI processes all audio as mono only. Always use mono microphones for optimal performance. Stereo microphones can introduce phase errors which may reduce detection accuracy.

DIY Microphone (Best Performance)

  • Recommended Capsule: PUI Audio AOM-5024L-HD-R

    • Characteristics:
      • High sensitivity (-24dB)
      • Low self-noise
      • Omnidirectional pattern
      • Weather-resistant when properly housed

  • Construction Requirements:

    • Cable: Shielded cable with braided copper or aluminum shield
      • Recommend: Mogami or Canare microphone cable
      • Cable length: Keep under 10m to minimize signal degradation
    • Connector: Quality metal 3.5mm TRS jack with proper shield connection
    • Power: Requires phantom power or bias voltage (can be supplied by many USB interfaces)

Pre-made Microphone Options

  • ** Lavalier (LAV) Microphones:**
    • Recommended models:
      • Boya USB BY-LM40 (frequently recommended in the community), in reality not very sensitive meaning audio captured is quite low volume
      • Clippy EM272 Mono Microphone - High quality omnidirectional microphone with excellent sensitivity and low self-noise Available here
      • Other omnidirectional lavalier microphones

Storage

  • Primary Storage: High Endurance MicroSD card

    • Capacity: 64GB or larger recommended
    • Speed Rating: V30 rated or higher to guarantee sufficient write performance
    • Type Specification: Look for cards explicitly marketed as:
      • "High Endurance"
      • "Surveillance Grade"
      • "Dashcam Rated"
    • Recommended Brands: SanDisk High Endurance, Samsung PRO Endurance, Western Digital Purple SC
    • Rationale: Standard MicroSD cards can wear out quickly due to constant read/write operations, leading to system failure. High Endurance cards are designed specifically for 24/7 recording applications.

  • Storage Utilization:

    • Keep MicroSD card usage below 80% for optimal performance and longevity
    • Consider external USB storage for clip exports if extensive archiving is planned

  • Alternative Storage Options:

    • SSD Boot (Raspberry Pi 4/5): Boot from USB SSD for improved reliability and performance
      • Requires Raspberry Pi OS 2021-01-11 or newer
      • Significantly improves system responsiveness and storage reliability
    • NAS Storage: Configure clip export to a network share for centralized storage
      • Requires stable network connection
      • Reduces wear on local storage

Networking and Connectivity

  • Wired Ethernet Recommended:

    • More reliable than Wi-Fi for 24/7 deployment
    • Required for maximum web interface performance, especially when viewing spectrograms
    • Essential for reliable RTSP stream ingestion

  • Wi-Fi Considerations:

    • Use 5GHz networks when possible (less interference)
    • Position for strong signal (> -65dBm) if relying on Wi-Fi
    • Not recommended for critical deployments or multiple RTSP streams

  • Bandwidth Requirements:

    • Basic web interface usage: ~1Mbps
    • Live audio streaming via web interface: ~0.5Mbps per stream
    • RTSP ingestion: Dependent on stream quality (typically 0.5-4Mbps per stream)
    • BirdWeather uploads: Minimal (~10KB per detection)

Power Supply Considerations

  • Power Supply - IMPORTANT:

    • Only use the official Raspberry Pi power supply matching your Pi model:
      • Pi 4/5: 5V/3A USB-C official power supply
      • Pi 3: 5V/2.5A micro-USB official power supply
    • Rationale for official PSU requirement:
      • Official Pi PSUs deliver 5.1V (instead of standard 5V) to compensate for voltage drop
      • Better EMI/RFI shielding than third-party options, reducing audio interference
      • Properly rated for the current demands of the Pi under load
      • Using non-official PSUs may lead to under-voltage warnings and unstable operation

  • Power Consumption Estimates:

    • Raspberry Pi 3B+: ~3W (idle) to ~5.5W (active analysis)
    • Raspberry Pi 4B: ~4W (idle) to ~6.5W (active analysis)
    • Raspberry Pi 5: ~5W (idle) to ~8W (active analysis)
    • Add ~0.5-1W for USB audio interface and microphone

Performance Optimization

Hardware Optimizations

  • Raspberry Pi Specific:

    • Overclocking: For Pi 4, consider modest overclocking to 1.8GHz for improved performance
      • Add to /boot/config.txt: 
        over_voltage=2
arm_freq=1800
    • Memory Split: Allocate minimum GPU memory (64MB) for headless installations
      • Add to /boot/config.txt: 
        gpu_mem=64

  • Storage Optimizations:

    • Consider disabling swap for SD card installations to reduce wear
    • For advanced users, move log files to RAM disk

Software Configurations for Hardware Constraints

  • Limited CPU Resources:

    • Reduce birdnet.overlap value (set to 0.0 for minimal CPU usage)
    • Limit the number of RTSP streams
    • Consider using birdnet.threads setting to reserve CPU resources for other tasks

  • Limited Memory Resources:

    • Disable thumbnail generation in the web interface
    • Use WAV format for audio exports (requires less processing)
    • Implement aggressive retention policies for audio clips

Additional Resources

  • Community Recommendations:

  • Testing Your Setup:

    • Use the built-in benchmark command to evaluate system performance: 
      birdnet benchmark
    • Test microphone sensitivity with known bird calls at measured distances
    • Compare detection rates across different hardware configurations

This document represents recommended configurations based on community experience. Individual requirements may vary based on specific monitoring goals, environmental conditions, and budget constraints.