New feature request: Text-to-speech (TTS) Solved!

[Svardsten53]

It would be nice to have a text-to-speech feature so you can hear what species have been detected. For me, it is interesting to be able to have detected species read out in headphones when you are out walking. The idea is to have the RPi computer connected to a power bank to make it portable and this I have tried and it works. The requirement then is that the text-to-speech function works offline. Another requirement is that it supports as many languages ​​as possible (Swedish for me). After a quick look on the internet I found the program Piper which seems to meet the requirements and is also open source and free. It is based on Python, which is already installed.
I have previously expressed this wish for species readings to the BirdWeather PUC and Merlin teams, but no feedback so far. It would be nice if BirdNET-Pi could be first and the solution shouldn't be too complicated.
The program Piper itself should be easy to install yourself with the language you want. What needs to be done in the code for BirdNET-Pi is the actual call to Piper of the species name each time a detection occurs.

https://github.com/rhasspy/piper

[Svardsten53]

I have managed to install Piper and it works when running from the command prompt.

I really would like to have some help from somebody knowing Python or PHP to implement the following string taken from the Piper instructions into the BirdNET-Pi code. Instead of 'Welcome to the world of speech synthesis!' the last detected species should be filled in, for example 'Blackbird'. Please note the species name must be surrounded by single quotation marks according to the Piper syntax. The updated string should then be sent to Piper as a command.

echo 'Welcome to the world of speech synthesis!' | piper \ --model en_US-lessac-medium \ --output_file welcome.wav

I have spent some hours with this trying to create a terminal command, but have zero Python skills and just get syntax errors or no response from Piper.

If I get this working I will of course write down instructions here.

Thanks!

[Svardsten53]

I guess I have found out how to write the command, but the call to Piper in the script doesn't work. I'm trying to find why.

[Svardsten53]

With a little help from ChatGPT with changing the code and writing instructions, I have solved the problem. I've tested it outdoors and it works just fine.

Installation Guide for Raspberry Pi

Step 1: Update and Reboot the System

1. Install Bookworm 64-bit Lite on your Raspberry Pi.
2. Open the terminal and run the following commands:

   sudo apt-get update
   sudo apt-get upgrade
   sudo reboot

Step 2: Install pip

• To install pip (Python Package Installer), run the command:

  sudo apt install python3-pip

Step 3: Install Piper

1. Install Piper TTS by running the following command:

   pip install piper-tts --break-system-packages

2. Reboot the system:

   sudo reboot

Step 4: Install BirdNET-Pi

• To install BirdNET-Pi, run the command:

  curl -s https://raw.githubusercontent.com/Nachtzuster/BirdNET-Pi/main/newinstaller.sh | bash

Step 5: Create a Directory for TTS

1. Create the directory /home/yourid/BirdSongs/TTS. Note: Do not place the directory inside BirdSongs, as it may be deleted if you run the "Clear ALL data" command.
2. Grant full permissions to the TTS directory:

   sudo chmod -R a+rwx /home/yourid/BirdSongs/TTS

Step 6: Download Piper Language Files

• Follow the instructions on the Piper GitHub page to download the necessary language files for your system.

Step 7: Modify the Code in reporting.py

1. Open the file home/yourid/BirdNET-Pi/scripts/utils/reporting.py.

2. Replace the existing code in the relevant section with the following code:

   def extract_detection(file: ParseFileName, detection: Detection):
       conf = get_settings()
       
       # Create a new filename based on detection data
       new_file_name = f'{detection.common_name_safe}-{detection.confidence_pct}-{file.root}.{conf["AUDIOFMT"]}'
       
       # Create a directory structure for extracted audio files
       new_dir = os.path.join(conf['EXTRACTED'], 'By_Date', f'{file.date}', f'{detection.common_name_safe}')
       new_file = os.path.join(new_dir, new_file_name)
       
       # Check if the file already exists
       if os.path.isfile(new_file):
           log.warning('Extraction exists. Moving on: %s', new_file)
       else:
           os.makedirs(new_dir, exist_ok=True)
           extract_safe(file.file_name, new_file, detection.start, detection.stop)
           spectrogram(new_file, detection.common_name, new_file.replace(os.path.expanduser('~/'), ''))
       
       # Use common_name_safe to create a TTS audio file with Piper
       command = f"""echo '{detection.common_name_safe}' | /home/yourid/.local/bin/piper --model /home/yourid/BirdNET-Pi/TTS/sv_SE-nst-medium.onnx --output_file /home/yourid/BirdNET-Pi/TTS/artnamn.wav"""
       
       # Run the command using subprocess
       result = subprocess.run(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
       
       # Check if the command ran successfully
       if result.returncode != 0:
           log.error(f"Error creating TTS: {result.stderr}")
       else:
           log.info(f"TTS audio created: {result.stdout}")
   
       # Specify the path to the audio file
       audio_file = "/home/yourid/BirdNET-Pi/TTS/artnamn.wav"
   
       # Play the file using aplay
       subprocess.run(["aplay", audio_file])
   
       return new_file

---

Audio Playback of Bird Names

For every new detection, an audio file with the bird's name will now be generated and played through connected headphones or speakers. If you want to send the audio to another device, you can use a Bluetooth transmitter connected to the Raspberry Pi’s audio output. Using the built-in Bluetooth functionality may be more complex and might not work as expected.

---

Notes:

• This guide assumes you have basic familiarity with the Raspberry Pi command line and how to navigate its file system.
• The TTS functionality requires Piper language models to be downloaded in advance.

---

[Svardsten53]

To achieve the fastest detection possible, the following values should be set in BirdNET-Pi's settings: Recording Length: 6
Extraction Length: 3