pebblab / field note

local speech analysis · open source

Phonetic
Mirror

Phonetic Mirror doesn't grade your voice. It gives you a way to see it.

Compare your voice with a reference. See the difference. Follow the patterns and notice subtle changes in your pronunciation.

Phonetic Mirror Vowels application showing reference and learner measurements, written feedback, and a vowel-space plot

reference learner

Once upon a time in the Kingdom of Academia…

Ella wanted to show her students the sounds they could not hear. An old laptop, a night janitor, a wizard, and a lobster on the Pi carried the idea somewhere improbable.

The open-source wizard holds a glowing tuning fork and a lobster token beside an old speech-analysis computer
Read the full fairy taleClose the story

He and She, and the Lobster on the Pi

Once upon a time, in the Kingdom of Academia…

A university campus with a brightly coloured STEM castle at its centre

There lived an ESL teacher named Cindy. But since her mom passed away, no one has called her Cindy anymore.

Everyone called her Ella thanks to her wicked twin stepsisters, Ziri and Olexa, in the STEM Department. They loved mocking Cindy and one of her students, Miguel, who also worked as a night janitor. The sisters would see them passing in the hallway and sneer: "Look, it's Él and Ella. He and She. What a pair of nobodies!" They knew that much Spanish, these smart girls.

Ella and Miguel pass Ziri and Olexa outside the STEM lab

Miguel was a quiet young man. He had a degree in engineering from his home country, but his English pronunciation kept him pushing a mop instead of designing bridges.

Ella's father, Linus, was the Chief Administrative Officer of the entire campus. Technically, he had the power to change things, but, a gentle, open-source kind of man, he was trapped in a proprietary marriage to Ziri and Olexa's mother, Dean Gwynn Googlynn.

Gwynn controlled the university's purse strings with an algorithm of such pure nepotism that poor Linus could do nothing but funnel the shiny grants into his stepdaughters' VR lab. All his Cindy got was a squeaky whiteboard marker and her mom's old laptop that ran phonetic analysis at the speed of continental drift.

Linus sits at his desk while Dean Gwynn Googlynn stands behind him

The Frozen Spell

Every day, Ella listened to her students. They had the grammar of scholars and the vocabulary of poets, but their pronunciation was trapped in a thicket of L1 interference.

Ella teaches pronunciation to a small university class

One evening, Ella had to stay in the adjunct office long after hours. Ziri and Olexa had dumped her in a ditch of data to label and organize into spreadsheets. "Girls' data should be pure, with no accent!" said Dean Googlynn and all three took off to a grant-funded gala, giggling like teenagers.

Dean Googlynn, Ziri and Olexa pose outside a STEM innovation gala

Ella's own work—a meticulous analysis of Miguel's latest recording—froze on the screen. The little pinwheel spun and spun. And spun. She had been trying to isolate the F2 transition of Miguel's /l/ sound. He was pronouncing "world" as "wor-ud". "If I could just show him the spectrogram drop-off, I know he would be able to fix it." But the laptop was drowning in the data. She thought of Miguel's face when he struggled to order coffee and of the thousands of students like him all over the world, who couldn't afford private tutors and accent coaches.

Miguel pauses during his night shift while Ella works at an old laptop

Exhausted and broken-hearted, Ella put her head on the non-responsive keyboard and wept.

Ella sleeps beside her notebook and frozen laptop

She fell asleep and dreamed of a biofeedback lab—Phonetic Mirror Castle—a place where the walls listened to you and drew a picture of your lips and tongue, and your pronunciation suddenly became visible at the very moment you spoke; where everyone's speech improved faster; where all the students felt safe; and that place was open to everyone, no matter the language.

An imagined open-air phonetic castle filled with learners and speech diagrams

The Wizard of Open Source

“I just want them to see the sounds they cannot hear. If only I had a carriage that could carry a load that heavy” she muttered in her sleep.

And then, as if from the fog of her childhood memories, a very distant relative appeared. He wasn't a godfather of fairy dust; he was the wizard of open source, a man who lived in a server farm in the decentralized Lower Lands where all code was free. He wore a boxy, slate-grey suit, a rigid menu-bar necktie, and a sweater vest knitted out of striking, chaotic bands of hot pink and electric yellow—the unmistakable colors of a wideband spectrogram. His thick, round spectacles looked like pitch-tracking nodes, and when he spoke, his voice didn't just hum; it resonated at a perfectly stable fundamental frequency of 100 Hertz, carrying a gentle Dutch accent. He had kind eyes and smelled faintly of soldering iron.

Ella meets the imagined wizard of open source holding a glowing tuning fork

"Oh, I see why this machine I once made for my second cousin's daughter was pinging me. You're still running my analysis engine on this old horse, ben je niet? Poor thing," he said softly, gently touching the laptop with the frozen screen.

Ella sniffled in her dream. " Your engine?"

The old man smiled, his gray eyes twinkling behind his glasses. "Yes, child. I am Praat, the Keeper of Phonetic Algorithms. But as the saying goes, you cannot teach an old dog a new trick. You need a younger, swifter companion. I'll take care of the old one."

He reached into his pink-and-yellow vest and pulled out a spectral tuning fork. It turned luminous and blow a spark " Pi is in Harness and the Claw is Open, " he muttered as if to himself. And suddenly the fork turned luminous and blew a spark into the dark room.

Ella blinked and open her eyes, puzzled and slightly disoriented. "A claw in the pie?"

Miguel, finishing his shift, stood in the doorway, a question in his eyes. He came closer and stretch out his hand. "Look what I found," In his open palm sat a small token—a flat, metallic charm with a tiny red lobster etched onto its surface.

"A... lov-ESS-ter," his accent's thickening on the unfamiliar word.

Miguel offers Ella a small mysterious lobster token in the dark classroom

"A WHAT?" Ella blinked again, wiping her eyes. She reached out for the token.

The moment her fingers closed around it, the token vibrated. It grew warmer and heavier until it expanded with a soft click. And then it turned into a sleek box. A toy-like computer box sat solidly in her palm. She saw a little booklet with a lobster on the side. In neat, tiny letters it read: " OpenClaw on Pi harness, designed in Austria. Installation guide."

Miguel crossed himself. Ella just stared.

The Building of the Carriage

This is the part of the fairy tale that usually gets skipped with a wave of a wand. But here, in Ella's world, this was the magic. Miguel looked at the little box with a mixture of awe and professional curiosity. He took the tiny computer from her, plugged it into the dusty ethernet port in the wall and an old monitor, and sat at the scuffed desk with Ella.

Ella and Miguel work together at a computer connected to a small lobster-marked box

Step 1: The Carriage Assembly

The computer blinked and hummed to life almost silently. "Great design," Miguel noted. Like a real engineer, he gave the little machine a clean, lightweight setup and kept only what they needed for sound analysis and scripting. No bloatware.

Ella's Journal: "Miguel says this makes the carriage lighter. It uses less RAM to think about windows and more to think about sound . He's impressed. He keeps muttering 'lov-ESS-ter' under his breath and shaking his head."

Step 2: Teaching Lobster to Listen

Miguel gathered the tools they needed. The screen flooded with lines of code.

  • Praat checked
  • Python checked
  • FFmpeg checked

"We are at your service," Miguel finally said.

"So the Lobster is like a... a butler?" asked Ella.

"Sí," Miguel said. "It waits. When it sees a new file—your student's voice—it reads a skill file. The skill file says: 'Do this, then this, then this.' Like a recipe."

"And the recipe tells it to use Praat?"

"Exactamente."

Step 3: Ella Teaches Lobster About Vowels

Ella pulled out her battered notebook. It was filled with years of hand-drawn vowel charts and acoustic measurements scrawled in the margins of student assignments. She flipped to a clean page

"Okay," she said. "Here's what Lobster needs to know about vowels."

She drew a quick trapezoid—the vowel space.

"Every vowel has a 'shape'. The shape is made by where I put my tongue. Praat can 'see' the shape by measuring two numbers. We call them F1 and F2."

She explained the Praat process in detail.

Miguel nodded slowly. "So you want Lobster to get these two numbers from the student's voice?"

"Correct. Then compare the two and tell the student how to adjust."

Miguel cracked his knuckles. "Okay. I can make Lobster do that."

Ella dictated from memory. She had done this manually a thousand times on her frozen laptop: Read file. Formant analysis. Burg algorithm. Get value at time. Extract F1. Extract F2.

Miguel translated her words into code. Ella read over his shoulder. "This is it? This tiny thing does what took me twenty minutes of clicking?"

"Sí. Lobster runs it in a second. And then it reads the results—like opening a tiny scroll that just says: 543, 1621."

"And then the Lobster knows my tongue was too high and too far back," Ella whispered.

Step 4: Ella Gives Lobster a Music Lesson

Ella turned to a new page in her notebook.

"One more thing. It's not just the tongue. It's the 'music'."

She drew a wavy line rising at the end.

"When Miguel asks a question in English, his voice should go up at the end. But he doesn't. He stays flat. Not because he can't—because he's spent years trying not to be noticed."

Miguel looked at the floor. He knew she was right.

"So I need the Lobster to measure the mean pitch—the average musical note of the voice. Praat calls it F0. The fundamental frequency. Praat can trace it over time. I don't want one voice to imitate another person's absolute pitch. I want the Lobster to show the contour: where the voice rises, where it falls, and how the melody changes."

Miguel nodded and added another part to the recipe.

"So the result now has two lines. First line: F1 and F2 from the vowel script. Second line: the mean pitch from the melody script. And the Lobster reads both."

Ella stared at the screen. "So the Lobster will know the 'shape' of the mouth AND the 'song' of the voice. From one recording?"

"Sí. A complete portrait."

Step 5: The Glass Slipper

Ella insisted on one final piece. "We can't just throw all this away after every lesson. We need to see progress. Every Miguel. Every vowel. Every rise and fall of every voice. I want to watch all of you getting better over time."

Miguel nodded. "Got it! A database. Like a journal that remembers forever."

He set up a simple progress log so each session could be saved and viewed later.

"Now," Miguel said, pointing at the screen, "you can look back at 'Miguel in October' and 'Miguel in April' and see how the picture changes."

Ella's Journal, final entry of the night: "I told Miguel what to measure. He taught the Lobster how to measure it. The carriage is built. Two spells live inside it now: one for the ghost of the tongue, one for the song of the soul. And Miguel—my Miguel—looked at the progress view and said, 'This is where I will watch myself become loud.' Tomorrow, we test it on a real voice."

That was their Glass Slipper. Not a shoe that fits a foot, but a tool that fits a voice—and two people who each knew exactly why the other needed to build it.

The small box beeped softly. A little lobster emoji appeared on the screen:

🦞

It was alive.

The Clock Strikes Midnight

Weeks passed. Ella's students stopped dreading pronunciation practice. They started bringing their own examples to feed into the machine. "What does my 'cat' look like today?" they'd ask, huddled around the screen, looking in the screen.

Miguel, the test pilot, saw his own formant plot of the word "engineer" overlap perfectly with the target for the first time. That afternoon, he walked into a job interview with a structural engineering firm and got the job. He didn't need the mop anymore.

Ella and Miguel present a phonetic analysis screen to students

The success was so loud it echoed down the stairs all the way through the Great Hall. Everyone heard it. Dean Googlynn heard it. Ziri and Olexa heard it.


One morning, the stepsisters squeezed into Ella's tiny classroom. They eyed the little computer box and the red lobster sticker. "Mmm... El-...Cindy, dear..." Ziri started, using her friendliest voice. "Our VR lab costs well... a lot, but our sets can't do... THAT," Olexa chimed in swallowing her pride like a mouthful of dry chalk.

Ziri and Olexa visit Ella beside a phonetic analysis display

"Could we... maybe... can we ask... Mmm... Miguel to teach OpenClaw to process our data too? Mommy said she's willing to... to pay him," the twins even stammered in unison.

Ella smiled. "You can try. But you'd better hurry up —today's his last day here. He starts his new job tomorrow."


The Happily Ever After

Cindy is now the head of the International Phonetic Institute . Her father, Linus, finally found his spine, divorced Dean Googlynn's proprietary grasp on the budget, and now works for the Open Labs Foundation , which funds OpenClaw labs for underprivileged schools worldwide.

Miguel? He drives a different kind of carriage now—a pickup truck for his construction-site visits. And he still swings by on Tuesdays to write a new Python scripts for "the girls."

Miguel speaks with colleagues at a construction site beside his pickup truck

Even Dean Googlynn asked him to call her Gwynn. Cindy coughed loudly to hide the chuckle, but quickly composed herself and chimed in, "We are all family here. Right, Miguel?"

Miguel sits at a computer while Gwynn, Ziri and Olexa look on

Later, after the trio was gone, Cindy and Miguel enjoy the quiet evening and remember the day when it all started.

Cindy and Miguel laugh and raise mugs together in the quiet office

"To Uncle Praat!" she says.

"To Señor Praat!" he echoes, and they clink their mugs.

"Remember Él and Ella?"

"And a claw in the pie?"

"Lovesster!"

and they both burst out laughing.

THE END

“A little off” is hard to practice.

Form a closure with the tongue tip at the alveolar ridge while lowering the velum and maintaining voicing.

Good instruction still needs a way back into the learner's own voice.

“I'm not a phonetician. So… what am I doing wrong?”

More than vowels.

Five focused desktop tools ask different questions of the sound.

  1. 01

    Vowels

    Compare F1 and F2 estimates, pitch, and intonation contours.

    formants · pitch · contour
  2. 02

    Stops

    Compare detected release timing, Voice Onset Time, and formant estimates near vowel onset.

    /p t k b d g/
  3. 03

    Fricatives

    Compare estimated spectral centre of gravity and mean intensity for detected frication.

    /s z f v ʃ/
  4. 04

    Nasals & Liquids

    Observe selected formant and intensity patterns in English nasals, /l/, and /r/.

    /m n ŋ l r/
  5. 05

    Progress Viewer

    Filter saved session logs and follow selected observations across attempts.

    patterns over time

Know your “partner.”

A reference is a chosen comparison point—not universal correctness or the one proper English voice. Speakers naturally differ.

Phonetic Mirror uses selected acoustic measurements to help make differences visible. Uses Praat for acoustic speech analysis.

  • F1 estimated first-formant frequency
  • F2 estimated second-formant frequency
  • F0 estimated fundamental frequency and pitch contour

F1 and F2 are acoustic resonance estimates, not literal measurements of tongue position. No single metric tells the whole story.

Example · illustrative metadata

A reference should have a context.

language / variety
US English
reference ID
US-01
word / target
see · /s/
reference type
curated
conditions
controlled
licensing
documented source

This example does not describe an existing recording. Metadata cannot rescue poor audio; it makes a deliberately selected reference more transparent.

A reference is a choice. Make the choice visible.

Patterns over scores.

Repeated attempts can reveal change across sessions, recurring patterns, and movement toward—or away from—a chosen reference. The point is not to compress a voice into one artificial score.

The progress line matters more than the finish line.

The picture is useful. The measurements behind it can be useful too.

Saved logs can be reviewed, filtered, compared, analyzed separately, and prepared for appropriate educational or research reuse.

Raw data is not automatically anonymous. When preparing a dataset for reuse, measurements can be separated from identifying information or deliberately pseudonymized.

A list of CSV progress logs for fricatives, nasals and liquids, vowels, and stops

Your voice doesn't need a trip to the cloud.

Open, yet private by design.

That is a design choice, not an absolute privacy or security guarantee. Inspect the source and use the software in an environment appropriate for your recordings and data.

Look inside.

Phonetic Mirror is an early public release for Linux, written in Python. Uses Praat for acoustic speech analysis. The source is available under the GNU GPL v3.

platform
Linux desktop
input
WAV audio
analysis
Praat, installed separately
license
GPL v3
View the source on GitHub

Praat is a separate project. Phonetic Mirror is not affiliated with or endorsed by the Praat team.

Where will it go?

That depends on what happens next — who uses it, who tests it, who teaches with it, who helps improve it, and whether there is enough support to keep building.

Maybe it stays a small open-source tool. Maybe it grows into an Open Phonetic Lab. This fairy tale does not have an ending yet — and every now and then, the right people can make something improbable happen.

An imagined future Open Phonetic Lab where learners and teachers gather around transparent speech displays outdoors