---
title: "Text to Speech in Python: 5 Ways (Offline, Free and API)"
url: https://apyhub.com/blog/text-to-speech-python
author: ApyHub
published: 2026-10-20T12:58:36.473787Z
---

# Text to Speech in Python: 5 Ways (Offline, Free and API)

# Text to Speech in Python: 5 Ways (Offline, Free and API)

## Introduction

To convert text to speech in Python, install a library such as **pyttsx3** (offline) or **gTTS** (saves an MP3), pass it your text, and play or save the audio. For natural-sounding offline voices, use **Piper**. For web apps, servers and documents, a **text to speech API** returns an MP3 from one HTTP request, with nothing to install.

| Method             | Internet needed          | Output              | Voice quality                  | Best for                            |
| ------------------ | ------------------------ | ------------------- | ------------------------------ | ----------------------------------- |
| pyttsx3            | No                       | WAV, or plays aloud | Robotic (your OS voices)       | Quick offline scripts               |
| Piper              | Only to download a voice | WAV                 | Natural (neural)               | Offline apps that should sound good |
| gTTS               | Yes                      | MP3                 | Clear, one voice per language  | Prototypes                          |
| edge-tts           | Yes                      | MP3                 | Natural (neural)               | Personal projects                   |
| Text to speech API | Yes                      | MP3                 | Natural, same on every machine | Web apps, servers, documents        |

## 1. pyttsx3: offline and built in

pyttsx3 uses the speech engine already on your computer (SAPI5 on Windows, NSSpeechSynthesizer on macOS, eSpeak on Linux), so nothing leaves your machine.

```python
import pyttsx3  # pip install "pyttsx3==2.98"

engine = pyttsx3.init()
engine.save_to_file("Your order has shipped.", "order.wav")
engine.runAndWait()
```

**Good:** free, offline and private. **Watch out:** the voice sounds robotic and changes from one operating system to another. In our test on Linux, the latest version (2.99) ran `save_to_file()` without an error but never wrote the file, while 2.98 worked. On Linux servers, install eSpeak first (`sudo apt install espeak-ng`).

## 2. Piper: natural voices, still offline

[Piper](https://github.com/OHF-Voice/piper1-gpl) runs small neural voice models on your own CPU. After a one-time voice download (`python -m piper.download_voices en_US-lessac-medium`), everything stays offline.

```python
import wave
from piper import PiperVoice  # pip install piper-tts

voice = PiperVoice.load("en_US-lessac-medium.onnx")
with wave.open("order.wav", "wb") as f:
    voice.synthesize_wav("Your order has shipped.", f)
```

**Good:** natural voices, many languages, fast enough for a laptop or a Raspberry Pi. **Watch out:** you download and manage the voice models yourself, and the output is WAV.

## 3. gTTS: free MP3s from Google Translate's voice

```python
from gtts import gTTS  # pip install gTTS

gTTS("Your order has shipped.", lang="en").save("order.mp3")
```

**Good:** one line, MP3 output, many languages. **Watch out:** it uses an unofficial Google Translate endpoint, not the paid Google Cloud product, so there's no SLA and heavy use gets rate limited.

## 4. edge-tts: neural voices from Microsoft Edge

edge-tts uses the online service behind Edge's Read Aloud feature. It's the most natural of the free online options, and `edge-tts --list-voices` shows every voice available.

```bash
pip install edge-tts
edge-tts --voice en-US-AriaNeural --text "Your order has shipped." --write-media order.mp3
```

**Watch out:** like gTTS, it depends on an unofficial service that can change or stop working without notice. Fine for personal projects, risky for a product.

## 5. A text to speech API: one request, an MP3 back

When speech is part of a product (order notifications, narrated articles, accessibility features, voice replies from an AI agent), you want a supported service that sounds the same on every server. ApyHub's [Text to Speech API](https://apyhub.com/apyhub/service/convert-text-to-speech) takes text and returns a link to an MP3:

```python
import os, requests

r = requests.post(
    "https://api.eu.apyhub.com/apyhub/convert-text-to-speech/text/link",
    headers={"apy-token": os.environ["APY_TOKEN"]},
    json={"text": "Your order has shipped.", "gender": "female"},
)
mp3_url = r.json()["data"]  # signed link, valid for 15 minutes
```

**What you get:**

* **Up to 75,000 characters per request** (roughly 12,000 words), so a long article is a single call.
* **A male or female voice**, set with `gender`.
* **Documents as input.** Send a PDF or Word file URL instead of text and get an MP3 of its contents. See [PDF to speech: turning documents into audio](https://apyhub.com/blog/pdf-to-speech-turning-documents-into-audio-with-one-api-call).

**What it costs:** 750 atoms per request, whatever the length. The free plan covers 4 requests a month to try it. On paid plans a request works out to about $0.06–$0.12, so narrating a 1,500-word article costs a few cents. See [pricing](https://apyhub.com/pricing).

**Watch out:** there are two voices. If you need dozens of voices, voice cloning or fine pronunciation control, Google Cloud Text-to-Speech, Amazon Polly, Azure Speech, OpenAI or ElevenLabs offer more, with more setup and their own pricing.

[**Try the Text to Speech API free →**](https://apyhub.com/apyhub/service/convert-text-to-speech)

## Which option should you use?

1. **Offline, and the voice doesn't matter?** pyttsx3.
2. **Offline, and it should sound natural?** Piper.
3. **A quick script that needs an MP3?** gTTS.
4. **A personal project with natural voices?** edge-tts.
5. **Part of a product, on servers, or reading documents?** A text to speech API.

## Common errors

* **pyttsx3 creates no file:** pin version 2.98 and call `runAndWait()` after `save_to_file()`.
* **pyttsx3 `OSError` on Linux:** eSpeak is missing (`sudo apt install espeak-ng`).
* **An empty 0-byte MP3 from gTTS or edge-tts:** both create the file before connecting, so a network failure leaves it empty. Check your connection or firewall.
* **gTTS `429 Too Many Requests`:** you hit the unofficial rate limit. Cache audio you've already generated, or move to an API.
* **API returns 400:** the body needs both `text` and `gender` (`male` or `female`). See our guide to [400 Bad Request errors](https://apyhub.com/blog/400-bad-request).

## Let an AI agent speak

The Text to Speech API is also available through [ApyHub MCP](https://apyhub.com/mcp), so Claude, ChatGPT, Cursor or Mistral Le Chat can call it directly: "turn this summary into an MP3." Every API in the [ApyHub catalog](https://apyhub.com/catalog) works the same way, so an agent can summarize a document and narrate it in one conversation. To try the request before writing code, save it in [Voiden](https://voiden.md), a free, open-source API client.

## Conclusion

pyttsx3 is the quickest offline option and Piper the offline option that sounds natural. gTTS and edge-tts give you free MP3s from online voices, and a text to speech API gives you supported, consistent audio for products, servers and documents. Experiment with a library, and switch to an API once your users rely on the audio.

[**Start free on ApyHub, no card required →**](https://apyhub.com/)

## FAQ

### How do I convert text to speech in Python?

Install gTTS and run `gTTS("Hello", lang="en").save("hello.mp3")`, or use pyttsx3 to work offline.

### What is the best text to speech library for Python?

pyttsx3 for quick offline use, Piper for natural offline voices, gTTS for simple MP3s, edge-tts for natural free online voices. For production, use a text to speech API.

### What is the best offline text to speech for Python?

Piper. It runs neural voices locally and sounds far more natural than pyttsx3.

### How do I save text to speech as an MP3 in Python?

gTTS, edge-tts and text to speech APIs produce MP3 directly. pyttsx3 and Piper save WAV, which ffmpeg can convert.

### Is gTTS free?

The library is free, but it relies on an unofficial Google Translate endpoint with no guarantees and possible rate limits.

### Why does pyttsx3 not save a file?

On Linux, pyttsx3 2.99 can finish `save_to_file()` without writing anything. Install 2.98 and call `runAndWait()` afterwards.

### How much does a text to speech API cost?

It varies by provider. ApyHub charges 750 atoms per request of up to 75,000 characters, about $0.06–$0.12 on paid plans, with 4 free requests a month.

## About ApyHub

[ApyHub](https://apyhub.com/) is a curated API catalog and the trusted operational layer between your application and the services it depends on. One subscription, billed in atoms, covers the whole [catalog](https://apyhub.com/catalog) of over 1,500 endpoints and capabilities, and it keeps growing. Every service is evaluated before it is listed and carries machine-readable certification covering GDPR, SOC 2 and ISO 27001. Every endpoint is MCP-ready by default. ApyHub runs dedicated EU and US regions and serves more than 65,000 developer workspaces every month. Start on the free tier with no card, or [publish your own API on ApyHub](https://apyhub.com/api-provider).

## Sources

* [pyttsx3](https://pypi.org/project/pyttsx3/), [piper-tts](https://pypi.org/project/piper-tts/), [gTTS](https://pypi.org/project/gTTS/) and [edge-tts](https://pypi.org/project/edge-tts/) on PyPI
* [Piper on GitHub](https://github.com/OHF-Voice/piper1-gpl)
* ApyHub, [Text to Speech API documentation](https://apyhub.com/apyhub/service/convert-text-to-speech) and [pricing](https://apyhub.com/pricing)