Confirmed on the employer's own hiring board on Nov 22, 2026. First seen by Alion on Nov 18, 2026. Recho scores A on the Alion truth index.
Senior Speech Data Platform Engineer
Description
## In short
You'll build the data engine behind Recho's speech models: continuously sourcing, processing, validating and delivering high-quality data for STT, TTS, VAD and full-duplex voice systems.
This is a hands-on senior role. You'll independently design and implement the full pipeline-from crawling and ingestion to cleaning, annotation, quality monitoring, benchmarking and dataset delivery. Your mission is to turn speech data production from a series of one-off projects into a reliable, scalable and measurable system.
- **Team language:** English (Japanese not required, but you should deal with Japanese)
- **Salary:** ¥10M-¥20M per year, plus stock options
- **Location:** Tokyo (Hanzomon), hybrid. Candidates based outside Japan are welcome. We offer visa sponsorship and relocation support, and we're open to discussing fully remote arrangements.
## You might be a fit if you are…
- A senior data engineer who has built large-scale pipelines for audio, video, text or multimodal datasets
- A speech or ML engineer who enjoys building the data systems behind model development
- An engineer who has owned the full lifecycle of datasets-from collection and processing to quality evaluation and production delivery
- Someone who can turn an ambiguous request such as "we need more natural interruption data" into a concrete data specification, pipeline and validated dataset
## About the role
Speech models improve only when the data behind them improves.
At Recho, we develop STT, TTS, VAD, speech-to-speech and full-duplex conversational models. These systems require more than large volumes of audio. They need correctly licensed, well-structured and measurable datasets covering different speakers, languages, accents, acoustic conditions, interaction patterns and failure cases.
Your core mission is to build a production-grade speech data platform that supports two complementary modes of data creation.
The first is **continuous data production**. You'll build pipelines that proactively discover, ingest and process useful speech data even before a specific model team requests it. The system should continuously expand our coverage, detect gaps and produce reusable datasets for future training and evaluation.
The second is **request-driven data production**. When researchers need a specific type of data-for example overlapping speech, interruptions, backchannels, noisy telephone audio, a particular accent or a rare conversational pattern-you'll translate that need into a data specification and independently deliver the dataset.
You won't just operate pipelines. You'll define what good data means, measure whether the pipeline is producing it and determine whether a dataset actually improves the target models.
## What this role is *not*
- Not a role focused only on writing crawlers or downloading public datasets
- Not a manual annotation-operations or vendor-management position
- Not a role where model researchers provide complete specifications for every dataset
- Not primarily an STT, TTS or foundation-model research role, although you'll work closely with researchers and need to understand their models
- General cloud infrastructure and cluster operations are handled by separate infrastructure teams
## What you'll do
### 1. Build a continuous speech data production system
- Design pipelines that continuously discover, ingest and process speech, audio, video, transcripts and conversational data
- Build crawlers and connectors for public, licensed, partner-provided, customer-provided and internally generated data sources
- Track source provenance, licensing conditions, consent, usage restrictions and dataset lineage
### 2. Build end-to-end processing pipelines
- Implement scalable processing for audio extraction, resampling, channel normalization, format conversion and corruption detection
- Build pipelines for VAD, segmentation, speaker diarization, language identification, transcription and timestamp alignment
- Detect and filter silence, music, noise, repeated content, synthetic audio, low-quality recordings and other unwanted data
### 3. Produce data for STT, TTS, VAD and full-duplex voice
- Build STT datasets with reliable transcripts, word-level timestamps, speaker labels and acoustic metadata
- Build full-duplex conversational datasets containing turn-taking, interruptions, backchannels, overlap, hesitation, latency and response-timing signals
- Define taxonomies and annotation schemas for conversational behaviors that do not fit traditional single-turn speech datasets
### 4. Deliver targeted datasets from ambiguous requests
- Turn research and product needs into measurable dataset specifications
- Decide whether the best solution is crawling, internal data mining, synthetic generation, human annotation or a combination
- Deliver documented and versioned datasets with clear quality reports, known limitations and intended use cases
### 5. Build quality monitoring and data observability
- Define quality metrics for every stage of the pipeline
- Detect pipeline regressions, source drift, distribution shifts and unexpected changes in dataset composition
- Maintain traceability from a training sample back to its source and every transformation applied to it
### 6. Benchmark data quality and model impact
- Measure the effect of new datasets through controlled model experiments and downstream evaluations
- Establish dataset acceptance criteria and prevent low-quality data from silently entering training
- Prioritize data work based on model impact rather than dataset size alone
### 7. Set the technical foundation for the data team
- Define the architecture, interfaces and engineering standards for speech data production
- Create internal SDKs and tools that allow researchers to discover, inspect and consume datasets
- Mentor future data engineers as the team grows
We're happy to go into more detail during the intro call and interviews.
## Why this role
- **Build the data flywheel behind every speech model.** Your work will affect STT, TTS, VAD, speech-to-speech and full-duplex voice systems. A pipeline you build once can improve many models and products over time.
- **Solve a new kind of speech-data problem.** Traditional speech datasets focus on isolated utterances and clean turn-taking. Human conversation includes interruptions, overlap, backchannels, hesitation, response timing and incomplete speech. Building data for full-duplex voice requires new schemas, pipelines and quality standards.
- **Establish the foundation of a growing data team.** The company needs a durable data capability, not just more datasets. The architecture and operating model you establish will become the foundation for future engineers and researchers.
## Tech stack
- **Languages:** Python, SQL
- **Data processing:** PySpark, Ray, Dask, Polars, Apache Beam or similar
- **Workflow orchestration:** Airflow, Dagster, Prefect, Argo Workflows
- **Storage and formats:** S3-compatible object storage, Parquet, WebDataset, Iceberg or Delta Lake
- **Streaming and queues:** Kafka, Pub/Sub, SQS or similar
- **Speech processing:** FFmpeg, torchaudio, librosa, Whisper, NeMo, pyannote.audio
- **Data quality and observability:** Great Expectations, Grafana, Prometheus or internal tooling
- **Infrastructure:** Kubernetes, Docker, cloud batch systems
- **Dataset and experiment tracking:** DVC, LakeFS, MLflow, Weights & Biases or similar
## Our product
We build the Recho AI Voice Platform, which delivers voice conversations natural enough to be hard to tell apart from a human. We also deploy and operate Voice AI Agents for large enterprises.
Our edge is a platform for building and improving high-quality agents quickly. It captures industry know-how and turns user feedback into a fast, continuous improvement loop. Our agents now handle complex, large-scale contact center operations in finance, manufacturing and the public sector. In some deployments, they've resolved every call without handing off to a human agent.
## Hiring process
Intro call → Case assignment & first interview → Paid work trial* → Final interview → Offer meeting
*The work trial is 1-2 consecutive days at our Tokyo office, working on real tasks alongside the team, and it's paid. It's a chance for both sides to check fit and get a feel for the work.
- **If you're based outside Japan,** we'll agree on the format together. We won't ask you to travel to Japan just for this step.
- If your schedule or your current employer's side-work policy makes the trial difficult, just let us know.
- We may also ask for references.
## About Recho
Recho develops foundation models for customer-support AI agents and also deploys them for clients, all within one organization. Our core strength is voice AI technology we built entirely in-house, which resolves the industry's long-standing trade-off between quality and cost.
We've closed a Series A of approximately ¥700M and are rapidly rolling out to demanding production environments at major companies such as SBI Shinsei Bank and KDDI.
## Learn more
- [Corporate website](https://recho-ai.com/)
- [Team interviews (note, in Japanese)](https://note.com/recho_ai)
- [Company deck](https://recruit.recho-ai.com/)
- [Case study: Partnership with KDDI and Altius Link (Japanese)](https://prtimes.jp/main/html/rd/p/000000021.000159800.html)
- [Case study: SBI Shinsei Bank (Japanese)](https://mainichi.jp/articles/20260130/k00/00m/020/275000c)
- [2-minute voice demo](https://recruit.recho-ai.com/#2f276fb799f480d5bba7d5784d6ad31e) - give it a listen!
Requirements
## In short
You'll build the data engine behind Recho's speech models: continuously sourcing, processing, validating and delivering high-quality data for STT, TTS, VAD and full-duplex voice systems.
This is a hands-on senior role. You'll independently design and implement the full pipeline-from crawling and ingestion to cleaning, annotation, quality monitoring, benchmarking and dataset delivery. Your mission is to turn speech data production from a series of one-off projects into a reliable, scalable and measurable system.
- **Team language:** English (Japanese not required, but you should deal with Japanese)
- **Salary:** ¥10M-¥20M per year, plus stock options
- **Location:** Tokyo (Hanzomon), hybrid. Candidates based outside Japan are welcome. We offer visa sponsorship and relocation support, and we're open to discussing fully remote arrangements.
## You might be a fit if you are…
- A senior data engineer who has built large-scale pipelines for audio, video, text or multimodal datasets
- A speech or ML engineer who enjoys building the data systems behind model development
- An engineer who has owned the full lifecycle of datasets-from collection and processing to quality evaluation and production delivery
- Someone who can turn an ambiguous request such as "we need more natural interruption data" into a concrete data specification, pipeline and validated dataset
## About the role
Speech models improve only when the data behind them improves.
At Recho, we develop STT, TTS, VAD, speech-to-speech and full-duplex conversational models. These systems require more than large volumes of audio. They need correctly licensed, well-structured and measurable datasets covering different speakers, languages, accents, acoustic conditions, interaction patterns and failure cases.
Your core mission is to build a production-grade speech data platform that supports two complementary modes of data creation.
The first is **continuous data production**. You'll build pipelines that proactively discover, ingest and process useful speech data even before a specific model team requests it. The system should continuously expand our coverage, detect gaps and produce reusable datasets for future training and evaluation.
The second is **request-driven data production**. When researchers need a specific type of data-for example overlapping speech, interruptions, backchannels, noisy telephone audio, a particular accent or a rare conversational pattern-you'll translate that need into a data specification and independently deliver the dataset.
You won't just operate pipelines. You'll define what good data means, measure whether the pipeline is producing it and determine whether a dataset actually improves the target models.
## What this role is *not*
- Not a role focused only on writing crawlers or downloading public datasets
- Not a manual annotation-operations or vendor-management position
- Not a role where model researchers provide complete specifications for every dataset
- Not primarily an STT, TTS or foundation-model research role, although you'll work closely with researchers and need to understand their models
- General cloud infrastructure and cluster operations are handled by separate infrastructure teams
## What you'll do
### 1. Build a continuous speech data production system
- Design pipelines that continuously discover, ingest and process speech, audio, video, transcripts and conversational data
- Build crawlers and connectors for public, licensed, partner-provided, customer-provided and internally generated data sources
- Track source provenance, licensing conditions, consent, usage restrictions and dataset lineage
### 2. Build end-to-end processing pipelines
- Implement scalable processing for audio extraction, resampling, channel normalization, format conversion and corruption detection
- Build pipelines for VAD, segmentation, speaker diarization, language identification, transcription and timestamp alignment
- Detect and filter silence, music, noise, repeated content, synthetic audio, low-quality recordings and other unwanted data
### 3. Produce data for STT, TTS, VAD and full-duplex voice
- Build STT datasets with reliable transcripts, word-level timestamps, speaker labels and acoustic metadata
- Build full-duplex conversational datasets containing turn-taking, interruptions, backchannels, overlap, hesitation, latency and response-timing signals
- Define taxonomies and annotation schemas for conversational behaviors that do not fit traditional single-turn speech datasets
### 4. Deliver targeted datasets from ambiguous requests
- Turn research and product needs into measurable dataset specifications
- Decide whether the best solution is crawling, internal data mining, synthetic generation, human annotation or a combination
- Deliver documented and versioned datasets with clear quality reports, known limitations and intended use cases
### 5. Build quality monitoring and data observability
- Define quality metrics for every stage of the pipeline
- Detect pipeline regressions, source drift, distribution shifts and unexpected changes in dataset composition
- Maintain traceability from a training sample back to its source and every transformation applied to it
### 6. Benchmark data quality and model impact
- Measure the effect of new datasets through controlled model experiments and downstream evaluations
- Establish dataset acceptance criteria and prevent low-quality data from silently entering training
- Prioritize data work based on model impact rather than dataset size alone
### 7. Set the technical foundation for the data team
- Define the architecture, interfaces and engineering standards for speech data production
- Create internal SDKs and tools that allow researchers to discover, inspect and consume datasets
- Mentor future data engineers as the team grows
We're happy to go into more detail during the intro call and interviews.
## Why this role
- **Build the data flywheel behind every speech model.** Your work will affect STT, TTS, VAD, speech-to-speech and full-duplex voice systems. A pipeline you build once can improve many models and products over time.
- **Solve a new kind of speech-data problem.** Traditional speech datasets focus on isolated utterances and clean turn-taking. Human conversation includes interruptions, overlap, backchannels, hesitation, response timing and incomplete speech. Building data for full-duplex voice requires new schemas, pipelines and quality standards.
- **Establish the foundation of a growing data team.** The company needs a durable data capability, not just more datasets. The architecture and operating model you establish will become the foundation for future engineers and researchers.
## Tech stack
- **Languages:** Python, SQL
- **Data processing:** PySpark, Ray, Dask, Polars, Apache Beam or similar
- **Workflow orchestration:** Airflow, Dagster, Prefect, Argo Workflows
- **Storage and formats:** S3-compatible object storage, Parquet, WebDataset, Iceberg or Delta Lake
- **Streaming and queues:** Kafka, Pub/Sub, SQS or similar
- **Speech processing:** FFmpeg, torchaudio, librosa, Whisper, NeMo, pyannote.audio
- **Data quality and observability:** Great Expectations, Grafana, Prometheus or internal tooling
- **Infrastructure:** Kubernetes, Docker, cloud batch systems
- **Dataset and experiment tracking:** DVC, LakeFS, MLflow, Weights & Biases or similar
## Our product
We build the Recho AI Voice Platform, which delivers voice conversations natural enough to be hard to tell apart from a human. We also deploy and operate Voice AI Agents for large enterprises.
Our edge is a platform for building and improving high-quality agents quickly. It captures industry know-how and turns user feedback into a fast, continuous improvement loop. Our agents now handle complex, large-scale contact center operations in finance, manufacturing and the public sector. In some deployments, they've resolved every call without handing off to a human agent.
## Hiring process
Intro call → Case assignment & first interview → Paid work trial* → Final interview → Offer meeting
*The work trial is 1-2 consecutive days at our Tokyo office, working on real tasks alongside the team, and it's paid. It's a chance for both sides to check fit and get a feel for the work.
- **If you're based outside Japan,** we'll agree on the format together. We won't ask you to travel to Japan just for this step.
- If your schedule or your current employer's side-work policy makes the trial difficult, just let us know.
- We may also ask for references.
## About Recho
Recho develops foundation models for customer-support AI agents and also deploys them for clients, all within one organization. Our core strength is voice AI technology we built entirely in-house, which resolves the industry's long-standing trade-off between quality and cost.
We've closed a Series A of approximately ¥700M and are rapidly rolling out to demanding production environments at major companies such as SBI Shinsei Bank and KDDI.
## Learn more
- [Corporate website](https://recho-ai.com/)
- [Team interviews (note, in Japanese)](https://note.com/recho_ai)
- [Company deck](https://recruit.recho-ai.com/)
- [Case study: Partnership with KDDI and Altius Link (Japanese)](https://prtimes.jp/main/html/rd/p/000000021.000159800.html)
- [Case study: SBI Shinsei Bank (Japanese)](https://mainichi.jp/articles/20260130/k00/00m/020/275000c)
- [2-minute voice demo](https://recruit.recho-ai.com/#2f276fb799f480d5bba7d5784d6ad31e) - give it a listen!
Preferred Experiences
## Nice-to-haves
### Speech and conversational data
- Experience building datasets for ASR/STT, TTS, VAD, speaker diarization or speech-to-speech models
- Understanding of audio sampling, codecs, channels, loudness, SNR and acoustic characteristics
- Experience with transcript alignment, speaker attribution and word-level timestamps
- Experience with conversational data containing overlap, interruption, backchannels or other full-duplex behavior
- Knowledge of phonetics, linguistics, prosody or speech annotation
### Data acquisition and governance
- Experience building and maintaining large-scale web crawlers
- Experience handling copyright, data licensing, consent, privacy and data-retention requirements
- Experience detecting and removing PII from audio and text
- Experience integrating customer, partner or vendor-provided datasets
- Experience designing human-annotation workflows and quality-control processes
### Data quality and evaluation
- Experience defining dataset benchmarks and acceptance criteria
- Experience measuring the causal impact of datasets on model performance
- Familiarity with audio-quality, intelligibility, speaker-similarity or conversational metrics
- Experience building dataset search, visualization or inspection tools
- Experience identifying dataset bias and coverage gaps
### Platform and technical leadership
- Experience designing internal data platforms, dataset SDKs or feature platforms
- Experience operating petabyte-scale or similarly complex data systems
- Experience setting technical roadmaps, architecture standards and review practices
- Open-source contributions to data, audio or ML infrastructure projects
- Experience building or leading an early-stage data team
### Languages
- Japanese or Chinese
- Experience working with multilingual or code-switching speech data
We are looking for
## What we look for
- **Hands-on:** You don't stop after drawing the architecture. You implement the critical pipelines, inspect the actual data and debug problems at the sample level.
- **Independent:** You can start from an incomplete request, identify the missing questions, define success criteria and drive the work through delivery.
- **Data-driven:** You measure dataset quality and model impact. You don't assume that more data is automatically better data.
- **Systems-minded:** You turn recurring requests into reusable pipelines, interfaces and standards.
- **Curious about the data itself:** You're willing to listen to audio, inspect transcripts and understand why individual examples pass or fail.
- **Pragmatic:** You combine crawling, open datasets, internal data, synthetic generation and annotation based on the needs of the task.
- **Responsible:** You treat provenance, licensing, consent and privacy as part of the system design-not as paperwork added afterward.
- **Team-first:** You build systems that make researchers and future data engineers more productive, rather than keeping important knowledge inside personal scripts.
## Our values
These are the values our team relies on to deliver Recho's value as fast as possible.
- **Exceed Expectations:** We keep exceeding the expectations of customers and teammates by delivering both speed and quality.
- **Team-First:** Each of us takes professional ownership, and we maximize the team's results, not individual ones.
- **Continuous Feedback:** We engage seriously and speak candidly for the sake of results. The harder something is to say, the more we say it with honesty, respect and care.
- **Earn Trust:** Trust is the foundation of every collaboration and every challenge. We act ethically and keep personal and company matters separate. Trust takes time to build and a moment to lose.
Working Conditions
Salary
## Salary
Compensation is based on your current compensation, skills and experience.
- **Annual salary:** ¥10M-¥20M
- **Salary review:** Once a year, based on your contribution and company performance
- **Stock option plan**
Location
WeWork Hanzomon (Hanzomon PREX North, 2-3-2 Kojimachi, Chiyoda-ku, Tokyo)
Hybrid: mostly in the office, with some remote work. For candidates based outside Japan, we’re open to discussing fully remote arrangements.
Job Type
Employment type: Full-time, permanent. Starting with a paid contract engagement first is also an option.
Work hours
## Working hours
- **Discretionary work system** (professional type under Japanese labor law)
- **Deemed working hours:** 9 hours/day
- Start and end times are up to you. As a general guideline, the team's core working window is 10:00-19:00, adjusted as work requires.
## Scope of duties and work location
This disclosure is required under Japanese labor law.
- **Duties:** Senior Speech Data Platform Engineers at hire; subject to change as determined by the company
- **Work location:** WeWork Hanzomon at hire (some remote work possible); subject to change as determined by the company
## Holidays & leave
Weekends, Japanese public holidays, year-end/New Year holidays (Dec 29-Jan 3), special leave, and paid annual leave (10 days in the first year)
Probation period
Probation period: 3 months, with no change in terms
Benefits
**Insurance**
- Full Japanese social insurance (health, pension, employment)
- Commuting allowance (up to ¥25,000/month)
**Office**
- WeWork Hanzomon, 2 minutes from Hanzomon Station, with a dedicated floor, focus booths and meeting rooms
- Free drinks, including coffee, soft drinks and alcohol
- High-quality office chairs
- MacBook Air and external monitor provided
**Productivity**
We build AI for our customers, so we work AI-native ourselves.
- **AI tools support**
- Engineering roles: Top-tier plans for Claude Code and Codex
- Business & PM roles: Team plans that let you share context across your team
- Amazon Business account for work-related books and equipment
**Team & communication**
Beyond project meetings and day-to-day conversations on Slack, we regularly set aside time for the whole company to share progress, discuss it together and get to know each other.
- **Monday Kickoff:** Every Monday, we share everything from work updates to little things going on in our lives.
- **Monthly All-Team Meeting:** On the first Friday of each month, each team shares its results from the previous month. Afterward, we get together over food.
- **All Hands:** Twice a year, we share company direction and our mid- to long-term vision, bringing the whole organization together.
**Smoking**
- Indoor smoking prohibited; outdoor smoking area available

