Python Speech & Audio Processing Specialist: Segmentation, Transcript Alignment & TextGrid Output

Project overview I am seeking a practical Python speech and audio-processing specialist to prepare two continuous English-language recordings and their supplied transcripts for forced alignment. The source materials consist of: one continuous English WAV recording of approximately 12 hours one continuous English WAV recording of approximately 5 hours and one complete corresponding transcript for each recording. The recordings and transcripts have not yet been divided into shorter utterance-level audio-and-text pairs. The selected freelancer will complete the majority of the technical preparation and first-pass forced-alignment workflow. The final package should be organized so that it can be reviewed independently if additional linguistic quality control is later required. No new transcription, editorial rewriting or speech-model training is requested. Required work The selected freelancer will: Review the source-material structure and recommend the simplest technically reliable workflow. Segment the two continuous recordings into practical alignment units using appropriate existing tools and automation. Map the supplied transcripts accurately to the resulting audio segments. Preserve the supplied transcripts as the authoritative text. ASR or AI-assisted tools may be used as timing or mapping aids, but they must not silently replace the supplied transcripts. Perform only the text normalization required for reliable forced alignment. Identify OOV terms and prepare or generate suitable pronunciation entries where required. Run the forced-alignment workflow and produce both word-level and phone-level TextGrid outputs. Identify and correct failed, dropped or clearly inaccurate alignments where technically practical. Preserve the original source timing so that every segment can be traced back to its start and end position in the original continuous recording. Deliver an organized first-pass package with a concise list of unresolved or questionable segments requiring later specialist review, if any. Existing tools such as Python, Praat, Montreal Forced Aligner, WhisperX, FFmpeg, voice-activity detection or technically equivalent methods may be used. The objective is not to build a new forced-alignment system or standalone commercial software product. The freelancer should use the simplest efficient workflow that produces the required outputs reliably. Expected deliverables The final delivery should include: segmented audio files matching segmented transcript files word-level TextGrid files phone-level TextGrid files source start and end timestamps for every segment a CSV or equivalent segment manifest OOV and pronunciation additions, where applicable a concise list of unresolved or problematic segments scripts, commands or configuration files used for the workflow and brief rerun instructions sufficient for another technical person to understand and reproduce the process. Scope boundaries This assignment does not require: retranscribing the complete recordings editorial rewriting or publication-quality transcript formatting manual inspection of every word and phone boundary manual gold-standard annotation of the full 17 hours development of a new forced-alignment algorithm development of a standalone software platform speech-model training or open-ended research and development. The focus is efficient technical preparation, reliable first-pass alignment, correction of clear failures and an organized handoff suitable for independent review. Preferred candidate background Strong applicants may describe themselves as: Python speech-processing specialists speech or audio engineers ASR engineers speech-data engineers computational linguists audio machine-learning engineers or developers experienced with long-form speech alignment. Direct Montreal Forced Aligner experience is preferred but is not the only acceptable background. Candidates with closely related forced-alignment experience may apply if they can clearly explain how they will produce valid word-level and phone-level TextGrid outputs. Relevant experience may include: Python audio processing long-form audio segmentation transcript-to-audio mapping voice-activity detection Praat TextGrid generation WhisperX FFmpeg forced alignment pronunciation dictionaries G2P OOV handling and speech-dataset preparation. Confidentiality The recordings and transcripts are proprietary. The selected freelancer must: use the materials only for this assignment; keep all project materials confidential; identify any cloud or external processing environment before using it; avoid uploading the files to consumer AI services or undisclosed third-party platforms and delete retained project materials after final acceptance if requested. Additional information and source files will be provided only to the selected freelancer after the required confidentiality arrangements are in place. Contract structure This is a fixed-price assignment. Applicants should provide one all-inclusive fixed price for the complete scope rather than open-ended hourly billing. The project may be organized into two milestones: Milestone 1 - Segmentation and first-pass alignment segmentation of both recordings; transcript mapping; alignment-specific normalization; initial OOV and pronunciation handling; first forced-alignment run; preliminary word- and phone-level TextGrids; and an initial exception list. Milestone 2 - Corrections and final handoff correction of failed or clearly inaccurate alignments final segmented files and transcript pairs final word- and phone-level TextGrids source-time manifest scripts or configuration files rerun instructions and a concise final exception list. Payment will be released against accepted milestone deliverables. Application questions: Please answer all five questions directly. Generic proposals that do not address them may not be considered. 1. Comparable experience Describe the closest comparable project you have personally completed. Include: approximate audio duration whether an existing transcript was supplied the tools you used and the final outputs you produced. 2. Proposed workflow In 200 words or fewer, explain how you would process two continuous English recordings of approximately 12 hours and 5 hours, with one complete supplied transcript for each. Explain how you would: segment the recordings map the transcripts preserve the supplied transcripts as the authoritative text and distinguish automated processing from active manual work. 3. Required technical outputs Can you produce both word-level and phone-level TextGrid files while preserving timestamps back to the original recordings? Identify the forced-alignment, segmentation and audio-processing tools you have personally used. Briefly explain how you would identify failed or clearly inaccurate alignments. 4. Confidentiality and processing environment How would you protect the confidentiality of the recordings and transcripts? State whether processing would occur: locally in a controlled cloud environment or through any third-party service. No undisclosed uploads or consumer AI services may be used. 5. Fixed price and completion time State: your all-inclusive fixed price anticipated elapsed completion time what your price includes and any material exclusions or assumptions. Please confirm that your proposed price includes the segmented files, matching transcript units, word- and phone-level TextGrids, source-time mapping, exception list and reproducibility materials.