DEV Community: Muhammad Usman SU92-BSAIM-F25-025

Switch-Point Prosody: Why AI Voices Break Where Pakistani Ads Switch Languages

Muhammad Usman SU92-BSAIM-F25-025 — Wed, 14 Oct 2026 05:50:33 +0000

A research proposal by Muhammad Usman, BS Artificial Intelligence, Superior University, Lahore — October 2026

The phenomenon

Walk through any commercial break on Pakistani television, or scroll through sponsored video ads on Instagram and TikTok, and a linguistic pattern repeats: the pitch is neither pure Urdu nor pure English. It is a deliberate mix — Urdu as the matrix language carrying the message, English insertions carrying prestige, urgency, and the call to action:

"Yeh phone best hai — limited offer, abhi order karo!"
"Naya smart TV sirf Rs. 85,000 me — free delivery ke saath!"

This is not sloppy speech. In Myers-Scotton's Matrix Language Frame model, bilingual speakers systematically embed one language's content morphemes inside another language's grammatical frame, and advertisers exploit this: English words signal modernity and trigger the purchase impulse, while Urdu carries trust and cultural proximity. The persuasive force of the ad lives exactly at the switch points — the emphatic stress on "best", the rising pitch into "order now!"

The problem

AI-generated voiceovers are now cheap enough for Pakistan's small businesses to use in video ads. But every practitioner who has tried knows the failure: feed a code-switched script to a TTS engine and the voice stumbles precisely where the ad needs it to soar. The Urdu words come out with English phonemes, the stress lands on the wrong syllable, and a flat pitch contour replaces the salesman's rise.

I verified this with a live demonstration: synthesizing "Yeh product best hai, order now" through a production English TTS voice, the Urdu function words receive English phonemization, stress defaults to English patterns, and the persuasive pitch rise into the call to action is absent.

What the literature says

Code-mixed speech synthesis is an established field — but for other language pairs. Sitaram and Black (CMU) built an early Hindi-English framework and found significant listener preference for systems that correctly identify and pronounce words in each language. Joshi and Garera (2023) reached production quality for Hindi-English e-commerce voice bots with a positive comparative MOS against a commercial baseline. Mandarin-English bilingual synthesis was demonstrated at Interspeech 2020, and an ICASSP 2020 system explicitly predicted log-F0 because phonetic representations alone lack the prosody code-switched synthesis needs.

Crucially, the prosodic disruption at switch points has been measured: a Chinese-English bilingual speech study found English words inside code-switched utterances showed nearly double the pitch variability of monolingual speech — average F0 standard deviation of 51.9 Hz versus 29.8 Hz. Speakers measurably alter their prosody around switch points, and a TTS system trained on monolingual prosody misrenders exactly those regions.

The gap: no published work addresses Urdu-English switch-point prosody, and none targets advertising persuasion — despite Urdu-English mixing being the everyday reality of hundreds of millions of speakers.

The proposed study

Phase 1 — Corpus: 50 code-switched Pakistani ad scripts, 5 native bilingual speakers, switch points annotated.
Phase 2 — Acoustic measurement: F0 contours via pYAAPT; compare 300ms windows around switch points against monolingual windows; synthesize the same scripts with commercial TTS engines and repeat.
Phase 3 — Perceptual evaluation: ABX listening tests (n ≥ 40): naturalness (MOS) and purchase intent ratings.

Expected contributions: the first acoustic characterization of Urdu-English switch-point prosody, a public annotated corpus, quantified human–synthetic prosody gaps, and a prosody-aware synthesis recipe evaluated on purchase intent — the metric advertisers actually care about.

Honest framing

This is a research proposal — the study has not yet been executed. Findings from Hindi-English and Mandarin-English systems may partially transfer, which would narrow but not eliminate the novelty. All literature claims are verifiable through the references in the full paper.

Do AI-Generated Video Advertisements Work for Pakistan's Small Businesses?

Muhammad Usman SU92-BSAIM-F25-025 — Tue, 13 Oct 2026 17:57:29 +0000

By Muhammad Usman, BS Artificial Intelligence, Superior University Lahore — October 2026

Status: This is a research proposal. The experiment described below has NOT been run yet. It is published here to invite collaboration and feedback.

The Problem

AI-generated video advertisements are exploding in popularity among small businesses: tools can now produce polished, voiceover-driven product ads in hours at a fraction of traditional production cost. Yet the experimental literature on whether these ads actually work is thin — and comes entirely from Western or English-language contexts:

A 21,000-consumer WhatsApp experiment (MIT/Harvard) reported a +9.4 percentage-point lift in click-through rate for AI-generated ads.
A June 2026 HKBU field experiment across 150+ ads found higher CTR and retention but lower conversions for AI-generated video ads.
A 2026 Syracuse/Ipsos study of 3,000 US consumers found AI ads 14% weaker in the short term and 17% weaker on brand equity than human-made ads.
Other work suggests CTR drops ~31.5% when AI involvement is disclosed to viewers.

These findings disagree with each other — and, crucially, none of them examined Urdu-language AI video ads for Pakistani SMEs distributed via WhatsApp, the channel much of Pakistan's small-business commerce runs on.

Research Questions

RQ1: Do Urdu-language AI-generated video ads beat human-produced ads on click-through and retention when distributed via WhatsApp broadcast lists?
RQ2: Does the conversion gap reported in Western studies replicate for low-price, everyday products in the Pakistani SME context?
RQ3: Does disclosing the ad's AI-generated nature reduce engagement among Pakistani viewers?
RQ4: Is synthesised Urdu voiceover the trust bottleneck — i.e., does viewer perception hinge on the voice rather than the visuals?

Proposed Methodology

Design: Randomised field experiment with 8–12 Lahore SMEs.
Stimuli: Matched AI-generated vs human-produced 15–30 second vertical ads for the same product, keeping product, offer, and price constant.
Distribution: Randomised WhatsApp broadcast split, with an additional AI-disclosure treatment arm.
Metrics: Attention (view-through/retention), conversion (inquiries, orders), a perception survey, and cost per ad.
Analysis: Pre-registered, with informed consent from all participants and anonymised aggregate reporting.

Expected Contributions

The first empirical evidence on AI video ad effectiveness in the Pakistani SME / Urdu-language / WhatsApp setting.
Practical guidance for Pakistani small businesses deciding whether to adopt AI video advertising.
A replicable, low-cost experimental protocol other researchers and practitioners can run.

Call for Collaboration

This is a research proposal inviting collaboration — the experiment is yet to be run. If you are a researcher, a Lahore-based SME owner, or an AI practitioner interested in this problem, I would love to hear from you.

Contact: su92-bsaim-f25-025@superior.edu.pk

Muhammad Usman: Emerging AI Engineer and Creative Technologist

Muhammad Usman SU92-BSAIM-F25-025 — Tue, 13 Oct 2026 17:56:46 +0000

Muhammad Usman: Emerging AI Engineer and Creative Technologist

Muhammad Usman — known professionally as Usman Kalera — is a BS Artificial Intelligence student at Superior University Lahore (Class of 2026, currently in his 3rd semester) and a creative technologist working at the intersection of generative AI, video production, and web development.

Background

From: Khushab, Punjab, Pakistan
Based in: Lahore, Pakistan
Education: BS Artificial Intelligence, Superior University Lahore (2026–present, 3rd semester)
Languages: English, Urdu, Punjabi

Skills

Programming: Python, C++
Generative AI: AI video generation, prompt engineering, AI art direction, Urdu/English text-to-speech voiceover
Web development: WordPress (Astra, Smart Slider 3, Contact Form 7, Rank Math SEO)

Work Highlights

AI-Generated Video Advertising

Rakhzaar product ad — a 22-second AI-generated video advertisement for a brand
Portfolio showreel — a 60-second AI video showreel featuring 3D illustrations
Meta-style hiring ad — a 20-second AI-generated hiring advertisement

Digital Growth Hub

A complete WordPress business website for a Lahore client: 7 pages, 5 posts, a homepage slider, a working contact system, and full SEO setup.

Portfolio Website

A published dark-theme personal portfolio website with 3D interactive elements — view it here.

Lahore Vlog Series

A multi-part cinematic vlog series on Lahore with English narration and Urdu subtitles.

Haveli — Aakhri Waar

An original Pakistani AI action-film concept, created as a proof of concept for AI-driven filmmaking in Pakistan.

Freelance & Business

Fiverr seller — usman0_kalera, offering AI video ads and reels
Usman Kalera (Google Business Profile) — video production and website design, Lahore
GitHub — Usmankalera-dev

Research & Community

Published "The Kalera Biradari: A Historical and Ethnographic Inquiry into a Punjabi Muslim Jat Gotra" (October 2026)
Selected for Pak Angels Cohort 12 — HEC-NCEAC/PEC Generative & Agentic AI training
Ultimate Research Bootcamp 2026, Batch 27 (WeMultiOmics)
Member of the Uraan Bano Shaheen youth volunteer network

Contact

Email: su92-bsaim-f25-025@superior.edu.pk
Portfolio: https://muse.ai/s/portfolio-website-x0xr67inxk9edxf

Open to collaboration on AI video projects, web development, and research.