Skip to content

Settings

BirdNET Live reuses one Settings screen across multiple workflows. The button opens the sections that are relevant to the screen you came from.

How Settings Scope Works

  • Opening Settings from Home shows the full screen.
  • Opening Settings from Live, Survey, Point Count, or File Analysis filters the screen to the relevant sections.

General

Theme

Choose Dark, Light, or System.

If Dynamic Color is enabled, BirdNET Live also tries to use your Android device's system palette. This only does something on supported Android devices; on iPhone and iPad the app keeps using the normal BirdNET Live theme, so turning the toggle on there changes nothing.

App Language

Sets the interface language.

Species Names

Controls the language used for species names. Follow app language uses the same language as the interface when that name is available.

Show scientific names

Shows scientific names below common names across the app.

Timestamp display

Controls how per-detection times appear in session review.

  • Relative shows the offset from the start of the recording, e.g. 00:12:34. Best for reviewing a single session and matching the spectrogram playhead.
  • Absolute shows the local clock time when the detection was captured, e.g. 08:42:17. Best for cross-referencing field notes, weather logs, or simultaneous recordings.

If a detection lands on a different calendar day from the session start (e.g. an overnight survey), the absolute time gains a +1d suffix so reviewers don't accidentally read tomorrow's dawn chorus as today's.

When Absolute is selected, an additional Show seconds in timestamps toggle appears. Disable it if you prefer the more compact 08:42 over 08:42:17 — useful when scanning long detection lists. Relative offsets always show seconds because reviewers need sub-minute precision to align with the spectrogram playhead.

Storage and exports always use UTC instants regardless of this setting, so the choice never affects the data — only the way it's displayed.

Audio

These controls appear in audio-driven live workflows.

Gain

Linear amplifier applied to incoming audio before it reaches the spectrogram and the classifier. Leave at 1.0× unless your input is consistently too quiet — for example a high-impedance lavalier mic on a phone, or a USB interface whose preamp is set too low. Pushing gain above 1.0 will not magically reveal calls that the mic never captured; it just rescales whatever the mic delivered, so loud nearby sounds may clip. Below 1.0 is useful in the rare case where a hot input is saturating the spectrogram.

High-pass filter (Hz)

Cuts low-frequency content before inference using a 24 dB/octave Butterworth filter — the slider value is the −3 dB cutoff. 0 Hz disables it. A 100–200 Hz cutoff strips wind, traffic rumble, and handling noise without touching most species; pushing toward 500–1000 Hz starts removing low whoots, owls, grouse, and bittern booms, so only go that high if you are deliberately ignoring those species in exchange for a much cleaner spectrogram in a noisy urban environment. The cutoff you pick should be visible as a sharp horizontal line on the live spectrogram.

Microphone

Lets you choose a specific input device or keep the System default. Your selection is remembered across app launches, so if you regularly use a USB or Bluetooth mic in the field you only need to pick it once. The same picker appears on the Survey setup screen.

Inference

Window duration

Controls the length of the analysis window.

Confidence threshold

Sets how conservative detections should be.

Sensitivity

Sigmoid steepness applied to the raw classifier output before the confidence threshold is checked. Higher values make the detector more permissive — fainter or more ambiguous calls cross the threshold, at the cost of more false positives. Lower values are stricter and only let confident detections through. The default of 1.0 matches the BirdNET reference. Try 1.25 if you suspect the model is missing distant calls; drop to 0.75 if you are flooded with low-quality detections of common species. Sensitivity is hot-applied: changing it mid-session takes effect on the next inference window.

Inference rate

Controls how frequently BirdNET runs inference.

Score pooling

Combines scores across recent inference windows so a single noisy window doesn't dominate the result. Off uses each window's raw probability — most reactive, noisiest. Average arithmetic-means the recent windows for the smoothest output. Max keeps the loudest peak per species, which is the most reactive smoothing mode and good for brief, sharp calls. LME (log-mean-exp, the default) is BirdNET's reference soft-maximum: it behaves like max when one window dominates and like average when several windows agree, which is usually what you want. Switching modes mid-session clears the rolling buffer so old logits don't leak into the new mode.

Pooling window count

Controls how many consecutive inference windows participate in score pooling. A larger value smooths each species' score over a longer time horizon, which suppresses spurious one-off detections — useful for steady, distant calls where you'd rather wait for a few corroborating windows before raising a detection. A smaller value reacts faster to brief vocalizations but lets through more noise. The default of 5 matches the value historically hard-coded into the model and is a sensible starting point for live use.

Spectrogram

FFT size

Controls frequency resolution in the spectrogram.

Color map

Choose Viridis, Magma, or Grayscale.

Duration (scroll speed)

Controls how much time is visible in the spectrogram window.

Frequency range

Sets the upper display frequency.

Log amplitude

Applies logarithmic scaling to the spectrogram for easier visual reading.

Announcements

This section controls whether BirdNET Live reads detections aloud through your headphones or the phone speaker while a session is recording. The whole feature is off by default because it changes the acoustic environment around the microphone — turning it on is a deliberate trade-off. There is no setup wizard: the verbosity × frequency pickers below are the entire setup, so you can tap a different preset at any time and immediately hear the difference. The intuition: in long surveys you can't keep glancing at the screen; a discreet voice in your ear means you can keep your eyes on the habitat and still know what was just heard.

Speak detections aloud (master toggle)

Off by default. When on, the app speaks each accepted detection using your device's built-in text-to-speech. Headphones are strongly recommended — using the phone speaker risks the announcement being picked up by the microphone and re-detected, so the app briefly mutes the recorder around each utterance to prevent that loop (see Mute mic while speaking below). When you have a screen reader (TalkBack on Android, VoiceOver on iOS) active the first time you launch the app, this toggle is enabled automatically as an accessibility default — you can turn it back off here at any time.

Verbosity preset

How much the app says about each detection. Minimal speaks just the species name (best for very long surveys where you only want the cue). Balanced is the default — short, varied phrases like "Robin", "Heard a Robin", "Robin again". Chatty adds a touch more context and is closer to having someone narrate alongside you. Custom appears automatically if you tweak the Advanced numerics by hand. The intuition: the same throttling settings can feel either too quiet or too noisy depending on phrasing — verbosity lets you keep the cadence and just dial the wordiness.

Frequency preset

How often the app is allowed to speak at all. Five steps from quietest to most talkative. Rare and Sparse wait a long time between announcements and cap the rate — well-suited to multi-hour surveys where you want a sense of activity without a running commentary. Normal is the default conversational cadence. Frequent shortens the gaps and lifts the cap; appropriate for short Live sessions or when you want closer-to-real-time feedback. Constant removes the startup delay entirely and lets the app speak on almost every detection cycle — useful for demos, accessibility, or whenever the gap before the first announcement on Frequent feels too long. Custom appears when you change the timing fields in Advanced. The intuition: this is the one knob that decides whether the app stays in the background or becomes a presence — tap a different preset and you'll hear the new cadence within the next detection cycle, no save button required.

Voice (speed and pitch)

Two sliders that adjust the platform TTS voice. Speed ranges 0.5×–1.5×; the default 1.0× is the platform "normal" pace. Pitch ranges 0.7×–1.3×. The intuition: a small reduction in pitch and a slight slowdown can make announcements much easier to parse outdoors with wind or moving water in the background; the Speak a sample button below previews three common bird names with the current settings so you can iterate without leaving the screen.

Advanced

A disclosure that exposes a handful of audio-routing toggles plus the trigger-mode picker. You generally do not need to open this — the verbosity and frequency presets above are the only knobs that matter day to day. The rate-limiting numerics (startup grace, minimum gap, max per minute, streak silence, recency reset) are bundled into the Frequency slider so there is one obvious place to dial cadence up or down.

  • Allow phone speaker — When off, announcements are silently skipped if no headphones or external speaker is connected. When on, the phone speaker is used as a fallback. Turn this on for casual listening at home; leave it off for fieldwork to guarantee no acoustic feedback into the microphone.
  • Mute mic while speaking — Replaces incoming audio with silence while the app speaks, so the speaker output cannot be picked up by the microphone and re-detected. Highly recommended (and the default). Only turn this off if your microphone is acoustically isolated from the phone speaker — for example a clip-on lapel mic on a different cable or a Bluetooth headset.
  • Lower other audio — Briefly reduces the volume of music or podcasts from other apps during the announcement and restores it afterwards. On by default. Off plays at full mix.
  • Cue tone before speaking — Plays a short, quiet tone before each utterance so your ear has a moment to switch from passive listening to attending to the voice. On by default. Particularly helpful when announcements are infrequent or when you have music playing in the background.
  • What to announce — Picks which detections are eligible for an announcement at all. Every detection (default) lets the throttling decide. First time per session announces a species only the first time it appears in the current session. Watchlist only limits announcements to species on your watchlist (useful for targeted survey work where you want to hear about your priority taxa and nothing else).

Recording

Mode

  • Full — save the whole recording
  • Detections only — save clips around detections
  • Off — no audio recording

Clip context

When Detections only is active, the app shows a single Clip context slider (0–5 s) that sets how much audio is preserved on both sides of each detection. Each clip is analysis window + 2 × clip context long, so with a 3 s analysis window and the default 1 s context the saved clip is 5 s. Setting the context to 2 s yields a 7 s clip (2 s pre-roll + 3 s analyzed audio + 2 s post-roll). Larger values give you more room for visual inspection or external review tools at the cost of disk space; 0 saves only the analyzed window itself.

Format

Choose WAV or FLAC.

Auto-start recording (Live mode only)

When enabled, Live mode begins recording as soon as the screen opens and the model finishes loading — no need to tap the microphone button. Useful for kiosk-style deployments, hands-free use (e.g. mounting the device in the field), or any workflow where the user already knows that opening Live always means "start now". Disabled by default so an accidental tap on the Live tile from the home screen does not silently begin a session. The auto-start fires only once per screen visit, so stopping a session and tapping the mic again still works as a manual restart.

Location

Use GPS

Use device GPS instead of manual coordinates.

Latitude / Longitude

Manual coordinates used when GPS is disabled.

Refresh GPS now

Forces a fresh location fix instead of reusing the last value the app cached. The intuition: GPS lookups are cached per-screen so a setup screen does not block waiting for a satellite fix on every open, but that cache can be miles out of date if you have driven to a new spot since the last session. Tap this when you have moved and want the geo-filter to use here, not where you started the morning. The current cached coordinates are shown in the subtitle so you can verify what the app thinks your location is. If GPS cannot get a fix within ~10 seconds, the app falls back to the OS-provided last-known location and warns you with a snackbar so you know the value is stale.

Download offline maps

Pre-caches OpenStreetMap tiles around your current GPS fix so the Survey live map and the exported HTML report still render a basemap when you're out of signal. The intuition: map tiles are streamed on demand by default, which is fine in town but useless in a forest valley with no cell service. Pick a radius (1, 5, 10, or 25 km) and the app downloads every tile in that square at zoom levels 12 through 16 — coarse enough to navigate, fine enough to read trails. The dialog shows an estimate (typically about 30 KB per tile) before you commit, and the request is rejected if it would exceed 50 MB to keep us a polite OpenStreetMap citizen. Downloads are paced under the 2 req/s tile-usage policy, and you can cancel mid-batch. Tiles land in the same on-disk cache that every map widget reads from, so a download done here is immediately visible everywhere — no extra wiring per feature.

Species filter

  • Off — no geographic filtering
  • Location filter — exclude species that fall below the geographic threshold
  • Location weighting — use the geo-model as an additional weighting signal

Geo-filter threshold

Appears when a location-based filter mode is active.

Export & Sync

Formats

Tick any combination of export formats — every save / share will bundle all the selected formats together inside a single ZIP. Pick a single format with no audio clips and no HTML report and you'll get a raw file (e.g. session.csv) instead of a ZIP, for backwards compatibility:

  • Raven Selection Table — for use in Cornell Raven Pro.
  • CSV — opens in any spreadsheet.
  • JSON — easiest for programmatic processing; carries the full per-session metadata.
  • GPX — track and waypoints for use in mapping tools (only meaningful when GPS was on).

The intuition: many workflows need more than one format at the same time — a CSV for the spreadsheet, a Raven table for the desktop reviewer, and a JSON for the analysis script. Untangling that with a single-format toggle used to mean exporting the same session three times. Now you tick all three once and they ride together in the ZIP.

Include audio files

Include saved audio alongside the exported tables or metadata when supported by the export workflow.

Include HTML report

When on, every export ZIP also contains a report.html file alongside the table, audio clips, and GPX. Open it in any web browser and you get a print-ready summary of the session: header card with date, location, observer, and totals; an interactive map of the GPS track and detection markers; a card per detection with the Cornell taxonomy thumbnail, names, score pill, your confirmation, any note you typed, and the original audio clip inline as a player; and the analysis settings used. The intuition: a CSV is great for analysis pipelines but useless for sharing with a non-technical collaborator or printing a quick field summary — the HTML report fills that gap with one tap. Species thumbnails and map tiles need a connection the first time the file is opened (they're fetched live from the BirdNET taxonomy API and OpenStreetMap), but everything else — text, layout, audio playback, links — works fully offline. Turn this off if you only need the raw data and want to keep the ZIP a few KB smaller.

Privacy

This section controls which third-party services BirdNET Live may contact on your behalf. Inference itself runs entirely on your device — these toggles only govern optional network features that enrich the experience. All three toggles are off by default on a fresh install; nothing reaches out until you say so. The intuition: each toggle is scoped to one concrete service and one concrete benefit, so you can opt into exactly what's useful to your workflow and nothing else.

Allow map tiles

Required for any interactive map in the app (the location picker, the Survey live map, the session map, and the map tiles inside the offline-tile downloader). When on, map widgets fetch raster tiles from the public OpenStreetMap servers; tile-coordinate requests reveal which area of the world you're viewing. When off, every map screen falls back to a placeholder card so the rest of the app still works without network leakage.

Allow place name lookup

When on, the app sends your recorded coordinates to OpenStreetMap's Nominatim service to resolve a short place name (e.g. "Berlin, Germany") that is shown next to the session in Session Library and Session Review. The intuition: numeric coordinates are precise but hard to scan when scrolling through a long list of sessions — a place name turns the list into something you can read at a glance. When off, sessions show the raw lat/lon only, and Nominatim is never contacted.

Allow weather lookup

When on, every saved session captures a one-shot snapshot of local conditions (temperature, precipitation, wind, cloud cover) at the recording coordinates and end time via Open-Meteo. The snapshot lands in Session Review under the location row and is mirrored into the JSON export, the per-session metadata block, and the HTML report. The intuition: weather is one of the strongest predictors of bird activity, and capturing it automatically — without you having to remember to check a separate app — turns every session into a more complete record. Open-Meteo is a free service and requires neither an account nor an API key. When off, no weather data is fetched or stored. Point Count and Survey setup also show a compact weather card near their location controls: it asks for this consent only when needed, previews the result as icon + temperature + wind once enabled, and reuses the same cached snapshot when the session is saved.

About

The About row opens the in-app About screen.

Danger Zone

Reset Onboarding

Shows the onboarding sequence again the next time the app launches.

Reset All Settings

Restores every preference on this screen to its default value. Sessions, recordings, voice memos, exports and downloaded map tiles are kept untouched — only the saved preferences (sliders, switches, picker choices) get wiped. The app closes after confirmation so the new defaults take effect on next launch.

Useful when you are not sure which slider you nudged that broke something, or when handing the device to someone else and you want a clean configuration without losing the data you collected.

Clear All Data

Opens a confirmation flow for permanently removing stored app data.

Workflow-Specific Parameters Outside Settings

Some parameters are configured inside their own setup screens rather than in the shared Settings screen.