Browse categories
Explore
Fiverr Pro
English
$
USD
I will fine tune whisper speech recognition models on your data
Mostafa
About this gig
I help you fine-tune Whisper speech recognition models on your own audio and transcript data to improve transcription accuracy for your specific use case, accent, domain, vocabulary, or audio conditions.
This gig can include model selection, dataset review, training setup, fine-tuning, transcription testing, and quantitative evaluation using evaluation metrics such as WER/CER. I can also help assess your audio and transcript quality, validate manifest files, identify data issues, and recommend fixes before training.
For more advanced projects, I can build or adapt an end-to-end fine-tuning app for audio clipping, transcript editing, manifest file management, audio/text quality checks, transcription scoring, model comparison, and fine-tuning/transcription job management.
Please contact me before ordering so I can review your dataset size, data format, target language, accuracy goals, and delivery requirements.
Get to know Mostafa
Mostafa
Ai Consultant
- FromUnited States
- Member sinceApr 2023
- Avg. response time1 hour
Languages
English
I hold a PhD in engineering and an MS in computer science, with strong experience in Python, machine learning, data science, automation, and full-stack software development. I build practical tools and AI systems end to end, from data processing and model development to APIs, dashboards, deployment, and workflow automation.
FAQ
Do I need to contact you before ordering?
Yes. Please contact me before ordering so I can review your dataset size, audio format, transcript format, language, and accuracy goals. This helps avoid ordering the wrong package.
What do you need from me to fine-tune a Whisper model?
I need audio files, matching transcripts, the target language, your expected use case, and any current transcription examples if available. A manifest file is helpful but not required for every project.
Can you work with messy or unprepared data?
Yes, but messy data may require extra dataset preparation. Poor transcripts, noisy audio, wrong timestamps, mixed languages, and inconsistent file formats can affect the fine-tuning result.
How much audio data do I need?
It depends on your goal. A small dataset can help test the process, but better results usually require enough clean audio/transcript pairs that match your real use case.
Can you improve accuracy for accents or domain-specific words?
Yes. Fine tuning can help with accents, names, custom vocabulary, technical terms, and recurring speech patterns, especially when the training data reflects the target use case.
Can you prepare my dataset for fine-tuning?
Yes. I can review audio quality, check transcript consistency, validate manifest files, clean formatting issues, and recommend fixes. Larger cleanup work may require a gig extra or custom quote.
Can you split or clip long audio files?
Yes. I can help with audio clipping and segmentation if your data needs to be split into smaller training samples. This may require an extra depending on the amount of audio.
Can you compare multiple models?
Yes. I can compare model outputs using transcription metrics such as WER and CER, then provide a clear recommendation based on accuracy, common errors, and practical use.
Can you build a full fine tuning app or workflow?
Yes. For advanced projects, I can build or adapt an app for audio clipping, transcript editing, manifest file management, audio/text quality checks, model comparison, and fine tuning/transcription job management.
Will my data stay private?
I treat client data as confidential and only use it for the agreed project. Please do not send sensitive data until we agree on the scope and handling requirements.
