2. TTS with Piper and OpenAI

In the previous lesson, we explored Espeak and Pico2Wave, two simple offline TTS engines on Raspberry Pi. Now, let’s take a big step forward and try two more advanced TTS options that offer higher voice quality and more flexibility:

Piper — a fast, neural network–based TTS engine that runs completely offline on Raspberry Pi.
OpenAI TTS — an online service that provides very natural and human-like voices, perfect for expressive speech.

These engines will make your Pironman 5 Pro MAX sound more realistic and lifelike. 🚀

1. Testing Piper

Piper is an offline neural TTS engine, meaning you don’t need an internet connection once the model is installed. It supports multiple languages and voices, making it a powerful option for embedded speech.

Run the program

cd ~/sunfounder-voice-assistant/examples
sudo python3 tts_piper.py

The first time you run it, the selected voice model will be downloaded automatically.
You should then hear the Pironman 5 Pro MAX say: Hello! I'm Piper TTS.
You can switch voices or languages by calling set_model() with a different model name.

Code

from sunfounder_voice_assistant.tts import Piper

tts = Piper()

# List supported languages
print(tts.available_countrys())

# List models for English (en_us)
print(tts.available_models('en_us'))

# Set a voice model (auto-download if not already present)
tts.set_model("en_US-amy-low")

# Say something
tts.say("Hello! I'm Piper TTS.")

Code explanation:

available_countrys() — Lists all supported languages.
available_models() — Lists available models for a specific language.
set_model() — Sets the voice model. If the model isn’t installed, it will download automatically.
say() — Converts text to speech and plays it immediately.

💡 Tip: Try different models to compare speed, clarity, and accents. Some models are lighter (faster), while others have higher fidelity.

2. Testing OpenAI TTS

Get and save your API Key

Go to OpenAI Platform and log in. On the API keys page, click Create new secret key.
Fill in the details (Owner, Name, Project, and permissions if needed), then click Create secret key.
Once the key is created, copy it right away — you won’t be able to see it again. If you lose it, you must generate a new one.
In your project folder (for example: /), create a file called secret.py:
```
cd ~/sunfounder-voice-assistant/examples
sudo nano secret.py
```

Paste your key into the file like this:

# secret.py
# Store secrets here. Never commit this file to Git.
OPENAI_API_KEY = "sk-xxx"

Run the program

cd ~/sunfounder-voice-assistant/examples
sudo python3 tts_openai.py

The program will connect to OpenAI’s TTS service, and the Pironman 5 Pro MAX will speak using natural, expressive voice output.
You can change voice styles and add instructions to control tone and expression (e.g., sad, dramatic, playful).
This makes OpenAI TTS ideal for interactive robots, storytelling, or educational assistants.

Code

from sunfounder_voice_assistant.tts import OpenAI_TTS
from secret import OPENAI_API_KEY

# Export your OpenAI_API_KEY before running the script
# export OPENAI_API_KEY="sk-proj-xxxxxx"

tts = OpenAI_TTS(api_key=OPENAI_API_KEY)
# tts.set_model('tts-1')
tts.set_voice('alloy')
tts.set_model('gpt-4o-mini-tts')

msg = "Hello! I'm OpenAI TTS."
print(f"Say: {msg}")
tts.say(msg)

msg = "with instructions, I can say word sadly"
instructions = "say it sadly"
print(f"Say: {msg}, with instructions: '{instructions}'")
tts.say(msg, instructions=instructions)

msg = "or say something dramaticly."
instructions = "say it dramaticly"
print(f"Say: {msg}, with instructions: '{instructions}'")
tts.say(msg, instructions=instructions)

Code explanation:

OpenAI_TTS() — Initializes the OpenAI TTS engine using your API key.
set_model() — Selects the TTS model (e.g., gpt-4o-mini-tts).
set_voice() — Chooses a specific voice (e.g., alloy).
say(text) — Converts the text to speech and plays it.
say(text, instructions=...) — Adds expressive tone instructions, allowing you to control the style of speech dynamically.

Example:

“say it sadly” → soft, emotional tone
“say it dramatically” → bold and expressive delivery
“say it excitedly” → enthusiastic tone

Troubleshooting

No module named ‘secret’

This means secret.py is not in the same folder as your Python file. Move secret.py into the same directory where you run the script, e.g.:
```
ls ~/
# Make sure you see both: secret.py and your .py file
```
OpenAI: Invalid API key / 401
- Check that you pasted the full key (starts with sk-) and there are no extra spaces/newlines.
- Ensure your code imports it correctly:
```
from secret import OPENAI_API_KEY
```
- Confirm network access on your Pi (try ping api.openai.com).
OpenAI: Quota exceeded / billing error
- You may need to add billing or increase quota in the OpenAI dashboard.
- Try again after resolving the account/billing issue.
Piper: tts.say() runs but no sound
- Make sure a voice model is actually present:
```
ls ~/.local/share/piper/voices
```
- Confirm your model name matches exactly in code:
```
tts.set_model("en_US-amy-low")
```
- Check the audio output device/volume on your Pi (alsamixer), and that speakers are connected and powered.
ALSA / sound device errors (e.g., “Audio device busy” or “No such file or directory”)
- Close other programs using audio.
- Reboot the Pi if the device stays busy.
- For HDMI vs. headphone jack output, select the correct device in Raspberry Pi OS audio settings.
Permission denied when running Python
- Try with sudo if your environment requires it:
```
sudo python3 tts_piper.py
```

Comparison of TTS Engines

Feature comparison: Espeak vs Pico2Wave vs Piper vs OpenAI TTS
Item	Espeak	Pico2Wave	Piper	OpenAI TTS
Runs on	Built-in on Raspberry Pi (offline)	Built-in on Raspberry Pi (offline)	Raspberry Pi / PC (offline, needs model)	Cloud (online, needs API key)
Voice quality	Robotic	More natural than Espeak	Natural (neural TTS)	Very natural / human-like
Controls	Speed, pitch, volume	Limited controls	Choose different voices/models	Choose model and voices
Languages	Many (quality varies)	Limited set	Many voices/languages available	Best in English (others vary by availability)
Latency / speed	Very fast	Fast	Real-time on Pi 4/5 with “low” models	Network-dependent (usually low latency)
Setup	Minimal	Minimal	Download `.onnx` + `.onnx.json` models	Create API key, install client
Best for	Quick tests, basic prompts	Slightly better offline voice	Local projects with better quality	Highest quality, rich voice options