Custom Micro Speaker Development: From Product Requirements to Mass Production
1. Why the Requirement Stage Decides the Programme
Short answer: Most custom speaker programmes fail at the requirement, not at tooling. A requirement written as component numbers rather than as product behaviour cannot be validated until a sample exists, and by then the change is expensive.
One point to establish before comparing them: configurations published as BOX are cavity modules, each of which requires a defined host cavity and cannot be mounted directly against a flat panel. The usual requirement arrives as a short list: a diameter, a power, an impedance, a resonance frequency. Those are component properties, and each of them is a means rather than an end. The reason the list is insufficient is that the same numbers can be satisfied by configurations that sound completely different, because the numbers do not describe the band the product operates in or the conditions under which it has to operate.
The cost of this shows up late. A programme that has specified a diameter and a power can validate neither until a sample is built, and if the sample then fails subjectively, the only remaining moves are to change the driver, change the cavity or renegotiate the requirement. All three are expensive at that point. A requirement that instead describes the frequency band, the listening distance, the noise floor and the available volume can be checked on paper before anything is built, which is what makes the rest of the sequence predictable.
There is a second, less obvious benefit to defining the requirement properly. A requirement that states the frequency band the product uses also tells the supplier where to measure. If the supplier's published figures are measured at a test point outside the band the product operates in, the figures will look adequate while the product does not. Choosing the test point is therefore part of the requirement, not a detail to be settled later.
2. Phase One: Define the Band, Not the Part
Short answer: The first deliverable is a frequency band, a listening distance and a noise floor. Those three convert an acoustic intention into something a driver can be specified against.
1. The frequency band the product actually uses. For voice announcement this is the speech band; for music it is wider and reaches lower; for a warning tone it may be a narrow region around one frequency. This single item decides more about the outcome than any component number.
2. The listening distance and the noise floor at that distance. These two set the level the output has to clear, and they are usually measured rather than estimated.
3. The available volume, stated as a cavity the structure can seal rather than as a total internal space. This converts directly into an installed resonance through FC = Fs × √(1 + Vas / Vb).
4. The drive voltage the supply can deliver, and the impedance the amplifier will drive. Together these set the reachable power through V = √(P × R), and they rule out configurations the rail cannot drive.
5. The duration and duty cycle at full level, which decides whether the rating that matters is a continuous figure or a short-duration one.
This phase produces no drawings and no tooling. What it produces is a requirement that can be checked before anything is built, which is the point. A band definition in particular is cheap to change at this stage and expensive to change after a sample has been evaluated, because everything downstream — driver selection, cavity design, test points — inherits it.
3. Phase Two: Prove the Band With a Sample, Not a Drawing
Short answer: The first sample exists to test the requirement, not the component. Build it around a known driver and the real cavity, and measure it at points inside the band the product uses.
The first sample has a specific job: to confirm that the band definition produces the intended result before the driver is committed. The cheapest way to do this is to use an existing driver whose behaviour is known, mount it in the cavity the structure will actually provide, and measure the assembled condition rather than the driver alone. A free-air measurement at this stage tells you almost nothing about what the product will do.
The measurement points should sit inside the band the product uses. For a voice product that means points across the speech band rather than a single conventional frequency; for a music product it means points that cover the range the content occupies. Measuring where the product operates is what makes the first sample's result predictive, and it also establishes the test points that will be used for acceptance later.
Where the requirement is a subjective one — a sound that is soft rather than harsh, or a level that feels right — the first sample should be evaluated by listening, with the competitor or reference unit measured alongside on the same day under the same conditions. A subjective requirement that is never converted into a measurable comparison cannot be validated, and will be renegotiated repeatedly.
4. Phase Three: The Tooling Decision
Short answer: Tooling is a decision about what will change, not about how many units. If a validated driver core can be reused, the tooling decision often disappears entirely.
Tooling is usually discussed as a volume question, which is the wrong frame. The relevant question is what has to change: if the programme can be met by a validated driver in the cavity the structure provides, no new tooling is needed and the decision is trivial. New tooling is only justified when the driver itself has to change, and at that point the question becomes whether the change can be achieved by adjusting parameters on an existing platform rather than by cutting a new one.
Table 1 lists the parameter changes that are typically available without new tooling, and what each one moves. These are the levers that make a custom programme inexpensive, and they are also the levers that most programmes discover late, after the tooling decision has already been made on the assumption that a custom part means a new part.
Table 1: Parameter changes available on an existing driver platform, and what each one moves.
Parameter | What it changes | Typical direction | Constraint it interacts with |
Diaphragm thickness | Free-air resonance; for a flat diaphragm the bending mode scales with thickness | Thinner lowers the resonance; thicker raises it | Raises mechanical fragility if taken too far; interacts with excursion |
Voice-coil layer count | Moving mass, and therefore resonance and sensitivity | More layers add mass, lowering resonance | Increases coil height, which can raise mounting height |
Voice-coil wire gauge | Moving mass and DC resistance | Thicker wire adds mass and lowers DC resistance | Interacts with impedance and the amplifier's load |
Dome material | High-frequency extension and stiffness | A stiffer dome extends the usable range and reduces edge | Adds mass, which lowers resonance; interacts with sensitivity |
Surround material and thickness | Resonance, linearity and low-frequency behaviour | A specific thickness can be chosen to place resonance where the product needs it | Interacts with cavity volume, since both set the installed resonance |
Magnet grade and secondary magnet | Sensitivity through magnetic circuit strength | Adding a secondary magnet raises sensitivity | Increases depth and mass; interacts with mounting height |
Port or cavity geometry | Installed resonance and low-frequency extension | A larger cavity lowers the installed resonance; a port follows a different route | Competes with structure, airflow and other internal volumes |
5. Phase Four: Acceptance Criteria and Change Control
Short answer: Acceptance criteria have to be stated with their measurement basis, and any later change has to be re-baselined against the same basis.
The acceptance criterion is where a development programme either becomes repeatable or does not. A sensitivity figure without its measurement basis — frequency, distance, drive voltage, cavity volume and mounting surface — cannot be checked at goods-in, because the receiving test will differ from the supplier's in at least one of those. The same applies to the resonance figure, and to any distortion or bandwidth claim.
The second part of this phase is change control. A driver that has been accepted will have some of its parameters adjusted over its production life, for reasons that have nothing to do with acoustics: a material substitution, a supplier change, a winding adjustment. Each of those has to be evaluated against the same measurement basis that the part was accepted on, because a change that is invisible on one basis can be significant on another. This is also why the band definition from phase one keeps its value long after the development is finished.
Project Case Study (Hongsheng) — A Band Definition Came Before a Drawing
The requirement came from an outdoor driving-test ground, where units are installed in roadside posts and announcements have to be understood at 3 m in daylight. The customer's specification included long-term outdoor exposure, moisture resistance and a temperature rise under direct sun. The starting point was a 40 mm 4 Ω 10 W full-range driver, and more than ten full-range 40 mm candidates had been evaluated without meeting the requirement.
What the requirement actually was, once it was stated properly, was a voice-band target of roughly 1 to 4 kHz at a 3 m distance with outdoor noise present. That definition immediately explained why repeated substitution had failed: general-purpose 40 mm drivers are full-range designs whose energy concentrates in the low midrange, so the share of output inside the speech band is low. At matched power the midrange level was insufficient, and no choice of full-range model changes that, because the constraint is the frequency band rather than the model. Defining the band was therefore the first deliverable, and it was what turned an unsolvable problem into a solvable one.
With the band defined, the driver was specified for the band rather than for the size. The diaphragm material was changed to a polyester film, raising level in the 1 to 4 kHz band. The free-air resonance was designed to 900 Hz, placing the driver's resonance peak in the midrange rather than below it — for announcement at distance, narrow-band midrange output serves intelligibility better than extended low-frequency output, which is the reverse of the usual expectation that a lower resonance is always better. Test points were then placed inside the speech band at 0.5, 0.8, 1.0 and 1.2 kHz rather than at a conventional single 2 kHz point, so the published figures describe the band the product uses.
The last three changes were parameter changes on the same 40 mm platform, with no change to the exterior or the structure: a secondary magnet added 2 dB, a shim adjustment reducing the magnetic gap from 0.25 mm to 0.18 mm added 1 dB, and a copper-clad aluminium voice coil added 3 dB, for 6 dB in total from magnetic circuit and moving-system optimisation alone.
Testing produced a sensitivity of 110 ± 3 dB average, measured at 10 cm at 0.1 W input. The 0.1 W basis is the correct one for a sensitivity figure and is stated explicitly so that it is not compared against power-driven measurements. The unit is rated 5 W with a 6 W maximum. Environmental qualification followed the product definition: temperature shock between −20 and +70 °C, static damp heat for 48 h, a 0.75 m drop test repeated ten times, a 48 h load test, plus lead-pull checks at 3.0 N horizontal and 2.0 N vertical, with the acceptance baseline that all testing returns the response within a stated deviation of the baseline. The magnet selected is N45H neodymium, whose material rating extends to 120 °C, while the product's specified operating range remains −20 to +70 °C.
The programme is in production with 15,000 units shipped. Alongside the acoustic work the rating was reduced from 10 W to 5 W and the unit price from 10 RMB to 6 RMB, so the customer saved 60,000 RMB across the shipped quantity while the measured level rose. The sequence that produced it is the transferable part: a band definition, a driver specified for that band, test points inside it, and parameter changes that required no new tooling.
Ten full-range drivers failed one requirement, because the constraint was the frequency band and not the model. Defining the band first, then adjusting three parameters on the same 40 mm platform, delivered 6 dB with no tooling change.
The distinction matters because it decides what a bench figure can be used for. Hongsheng states the cavity condition alongside each published configuration, so that a bench measurement and a product measurement can be compared on the same basis.
This sequence is also what Hongsheng applies on custom programmes: the frequency band is defined and checked against a published configuration before a structure is committed, so the tooling decision is made on evidence rather than on assumption.
6. What to Bring to the First Meeting
Short answer: Six items make the first conversation productive: the band, the distance, the noise floor, the volume the structure can seal, the supply voltage and the reference unit if there is one.
6. The frequency band the product uses, stated as a range rather than as a single point.
7. The listening distance and, if it is known, the noise floor at that distance.
8. The volume the structure can actually seal, and the dimensions it cannot give up.
9. The supply voltage and the impedance the amplifier will drive, which together set the reachable power.
10. A reference or competitor unit if one exists, so that a subjective requirement can be measured against something.
11. The expected production volume and timing, which decides whether a validated platform can be reused or a new one has to be cut.
7. FAQ: Custom Micro Speaker Development
Q1: How long does a custom micro speaker programme take?
The phases scale with what has to change. A programme that can be met by a validated driver in the cavity the structure provides needs no tooling and moves at sample speed. A programme that needs a new driver platform adds a tooling phase and a pilot run. The requirement phase in either case is short, and shortening it further is usually a false economy.
Q2: What is the single most common cause of failure?
A requirement written as component numbers rather than as product behaviour. The same diameter, power and impedance can be satisfied by configurations that sound completely different, so a programme can meet every number in its specification and still fail subjectively.
Q3: Do I need new tooling for a custom speaker?
Often not. Diaphragm thickness, voice-coil layer count and wire gauge, dome material, surround material, magnet grade and port geometry are all parameters that can be adjusted on an existing platform. In one programme, 6 dB of level was delivered by three such adjustments with no change to the driver's exterior.
Q4: Should I specify the resonance frequency?
Specify the band the product uses and the installed resonance you need, not the free-air value. Because a sealed cavity raises the installed resonance according to FC = Fs × √(1 + Vas / Vb), a free-air figure alone does not tell you what the product will do.
Q5: Why do the published figures not match my measurement?
Usually because the measurement basis differs. Frequency, distance, drive voltage, cavity volume and mounting surface all have to match. Where a catalogue states different cavity volumes for its sensitivity and resonance figures, even two figures from the same row may not describe the same condition.
Q6: How do I write an acceptance criterion that survives production?
State the figure with its full measurement basis, and re-baseline against the same basis whenever a parameter changes. A criterion without a basis cannot be checked at goods-in, and a change that is invisible on one basis can be significant on another.
Q7: My requirement is subjective. Can it be specified?
Yes, by converting it into a comparison. Measure a reference or competitor unit on the same day under the same conditions, and state the requirement as a relationship to that measurement rather than as an adjective. A subjective requirement that is never converted into a measurable comparison gets renegotiated repeatedly.
8. Summary: Define the Band, Then Build Toward It
A custom micro speaker programme is decided at the requirement stage, and the requirement that works is a description of what the product has to do rather than a list of component properties. The frequency band, the listening distance and the noise floor convert an acoustic intention into something a driver can be specified against; the available cavity volume and the supply voltage then rule out what cannot be built. Everything after that is verification: a sample that tests the requirement rather than the component, test points inside the band the product uses, and acceptance criteria stated with their measurement basis so they survive production.
The economics follow from the same logic. A programme whose requirement is defined in behavioural terms can usually be met by adjusting parameters on a validated platform, which needs no new tooling; a programme whose requirement is defined as a new part will get one. Hongsheng can evaluate a required frequency band against a published configuration before a structure is committed, and its engineering team can state which parameter has to move first when a target cannot be met as drawn.
Next step If you are starting a custom micro speaker programme, the shortest route to a configuration worth testing is to state five things: the space available for the driver and its cavity, the impedance the amplifier will drive, the power the rail can supply, the level required at the intended listening position, and the frequency band the product actually uses. With those specified, our engineering team can recommend a suitable configuration for evaluation, or state plainly which part of the acoustic design has to change first.
More in This Series
· Micro Speaker Customization: What Can Actually Be Changed, and What It Costs →https://www.hsdz-spk.com/news/585.html
· OEM Micro Speaker Development Process: From Specification to Mass Production → https://www.hsdz-spk.com/news/586.html