---
title: "Troubleshoot"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Troubleshoot}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  eval = FALSE
)
```

## Long Pause After Prediction

If you notice a significant pause after BirdNET has finished processing your audio file but before returning the results, this is likely due to data conversion overhead between Python and R when calling `as.data.frame()`. This is particularly noticeable when:

- Processing long audio files (>20 minutes)
- Using low confidence thresholds that generate many predictions (`min_confidence < 0.1`)
- Working with memory-constrained environments

### Solution: Use `write_predictions()`

For large prediction results, prefer `write_predictions()` to write results directly from Python to a file (CSV, Parquet, or NumPy) without crossing the R boundary. This avoids the memory overhead of converting large result sets to R data frames.

```{r}
model <- load_birdnet()
predictions <- predict(model, audio_path)

# Write directly from Python - no R conversion needed
write_predictions(predictions, "results.csv")
```

For tips on processing many files efficiently, see the
[Get started with birdnetR](birdnetR.html) vignette.