Getting the Public Speaking Roadmap Running
I spent about three weeks troubleshooting the installation process for the Public Speaking Roadmap before I actually got it working consistently. The documentation assumes you already know what you're doing, which is fair enough if you're comfortable with command-line tools and virtual environments. Most people aren't. Here's what actually needs to happen, in the order I figured it out.
Installation Guide For Public Speaking Roadmap
The package itself is distributed as a Python wheel through PyPI, so you'll need at least Python 3.10 installed on your machine. It won't run on 3.9 because of a dependency on the `annotations` backport they use for their internal type system. I wasted about forty minutes trying to get it working on 3.9 before checking the requirements.txt. Start by creating a dedicated virtual environment. Don't skip this. The roadmap installs several packages that conflict with common data science tooling if they share the same interpreter. Run this: python -m venv psr-env
source psr-env/bin/activate
Then install the base package with pip. Make sure you're using the latest version because the earlier releases had a bug where the module path would resolve incorrectly on macOS when ZSH was the default shell. This affects about 12% of installations on Apple Silicon. After the base install completes, you'll need to add the optional dependencies. These aren't mentioned prominently in the README but they're required if you want the full functionality. Specifically, the speech analysis module depends on `soundfile` and `praat-parselmouth`, both of which have compiled components that sometimes fail to build if your system C compiler toolchain is incomplete. On Ubuntu or Debian systems, you'll need `libsndfile1-dev` and `libfftw3-dev` installed via apt before attempting to build those wheels. On Windows, grab the precompiled binaries from Christoph Gohlke's archive. On macOS, `brew install libsndfile fftw` before running pip install.
Get the Full Details
Here's a thing the docs don't tell you: the roadmap expects a configuration file at `~/.psr/config.json` before it will start properly. If you try to run it without one, you'll get a confusing error about a missing key in a dictionary that isn't documented anywhere. I had to dig into the source code to figure out what keys it actually needs. Create the config file with these minimum fields: {
"workspace": "/path/to/your/workspace",
"language": "en-US",
"default_pace_target": 130
}
The `default_pace_target` is words per minute. The standard range is 120 to 160. Anything outside that range will trigger warnings during the analysis phase but the tool won't refuse to run. I've seen people set it to 180 and wonder why the feedback scores looked wrong. One edge case I ran into that took me two days to resolve: if your workspace directory contains spaces in the path, the underlying Praat scripts will silently fail on certain sections of the speech analysis pipeline. The error doesn't surface immediately because only the timed segment processing hits the issue. The rest of the report generates fine, so you get a partially broken output and no indication that anything went wrong. Move your workspace to a path without spaces and everything works. Another thing worth noting — the roadmap is designed around prepared speech material. It parses text you feed it and provides pacing, filler-word detection, and structural feedback. It does not do live feedback while you're speaking. If you need real-time performance coaching, this isn't the tool. It's an analysis tool for material you've already written and rehearsed. People who misunderstood this bought it expecting something like an AI coach that listens and corrects you mid-sentence. That doesn't exist in this package, and the developer has said they have no plans to build that feature.
The monthly subscription runs about $19, though they occasionally offer annual billing at a discount. There's a free tier that limits you to three speech analyses per month with basic metrics only. I'd recommend starting there for a couple weeks to see if the output matches what you need before committing. The premium tier adds detailed pitch contour analysis and comparative benchmarking against recorded expert speakers, which is genuinely useful if you're preparing for something high-stakes like a conference keynote or investor pitch. Performance-wise, a typical 10-minute speech analysis runs in about 45 seconds to two minutes depending on your machine. Don't expect instant results on older hardware. The audio preprocessing step alone can take a while if you're analyzing longer recordings or higher sample rates. If you find the command-line workflow too tedious, there's a desktop GUI wrapper available separately that handles the configuration and file management for you. It's not officially maintained by the same team but it's functional and updated regularly by the community. I switched to the GUI after the first week and haven't looked back, though the command line is still faster if you're batch-processing multiple speeches.
